Generative AI Art Generation: Mastering all the AI Tools - Midjourney, BW, DALL-E, SD, Runway, etc.

1. Welcome to AI Art Generation: The medium for artists have always changed over history, from cave paintings to sculptures, to architecture, to canvases, to digital art, to virtual reality. And now we are in the realm of A I. There are now dozens of AIR generation tools out there. And as an AI artist, you need to be familiar with all of the important ones. And that's what this course is all about. Welcome to AIR Generation, Mastering all the AIR Tools. We created the scores because we saw teachers making A generation horses, but only talk about Meat Journey, or only Dali, or only Runway. That's not the right way to think about it. Imagine you're learning to be a carpenter. You don't just learn how to use a hammer. Instead you learn how to use all the tools like a screwdriver, arrange, and so on. Different tools to different things. For example, Meat Journey is great for the realism but it's not a tool for, in painting. For that you can use Leonardo if you're doing image customization, stable diffusion using automatic 11. 11 is your tool. How about if you need to upscale, relight, or edit, then clip drop. Adobe Firefly is the way to go get my drift. This class is created for someone that wants to be a well rounded artist, that's who we are. I'm Katrina and I'm an artist and videographer that creates a still video content for tech firms using AI technologies. And I'm Henry. I teach AI technologies to people to help enhance their productivity and create automations. In this course, we deep dive into the 12 most used AIR generation tools. We create photorealistic images with mid journey. We do image to image generation with blue willow. We relight and reshape with clip drop. We do generative fills with Adobe firefly. We do out painting with Leonardo. We even apply a custom model using stable diffusion, automatic 11, 11 and much, much more. Now this is an 11 hour course, but trust me, by the end of this course, you will be a well rounded AI artist, will be competent in all of the popular AI art generation type tools. The class project is by far the most exciting thing that we've done. In fact, you will have three options. You can either make a logo for business, you can make a computerized version of yourself, or you can make a book cover. So what are you waiting for? Start the journey to become an artist Today, it's as easy as clicking this button. We'll see you inside. 2. AI Image Generator Apps Introduction: In this course, we will be covering a lot of AA tools and apps. It's easy to get confused, that's why I created this chitchat that will help you navigate through how this course here I've listed the major image generator platforms that will cover. Here we have Ali images, do Lexcams Studio, blue willow, mid journey, leonato, and Automatic 11, 11. Let me now tell you a little bit more about each platform for the model. Dali uses a proprietary model. Dali images is based on stable diffusion. Lexica Art is also stable diffusion based. They have their own proprietary model. Dream Studio is using stable diffusion. Blue Willow is stable diffusion based. They use their own algorithms to choose a specific stable diffusion model based on your prop. Then we have Mid Journey. They have their own model. Leonardo is also stable diffusion based. So they have a lot of stable diffusion models as well as a proprietary model. And in Automatic 11, 11 is where you can use any open source staple diffusion model. How easy it is to use each platform. Delhi Images I Lexica. It's pretty easy to use for Dream Studio, I would say. Not really because they have a pretty difficult user interface for blue willow and in journey it's pretty easy as long as you have Discord account. If you don't have that, you'll have to sign up for Leonardo is also somewhat easy. The thing with Lado, they have lots of settings and parameters which is great for more advanced users. And automatic 11, 11 is not beginner friendly, it's for advanced users. Then some platforms would be on their website, some will be on Discord like Blue Willow and Mid Journey. And for Automatic 11, 11, it's a web interface of stable diffusion which you can install locally on your computer or use a cloud server. Next, for some platforms, you would need to write a longer prompt to generate better images. However, for other platforms, that's not the case, and basic prompts work as well as longer prompt. Those platforms are Lexica and Mid Journey. If you write some basic prompts with these platforms, you will generate great images. Then I've listed some features that you can find on each platform. The first one is Image to Image Generation. This is where you can upload your image and generate based on this image. For Dali. You can do that for Lexica.org You can do Dream Studio? Yes. Blue willow, lado and Automatic 11, 11. Everything except images. Then we have in painting allows you to edit a specific part of your image basically by erasing and writing a new prop. Some of the platforms that have this feature are Dali, Leonardo, and Automatic 11 11 for out painting. Out painting is basically extending your image. The platforms that have it are Al Lexico art, blue willow, mid journey Leonardo and Automatic 11 11 for image quality. This is my personal evaluation. I found that with Lexico Art, you get high quality images mid journey, great images, Leonardo and Automatic 11, 11. You can also achieve really high quality images, but with Leonardo and Automatic 11, 11, they require some experience to get high quality images. So you need to know the settings that you work with. Then for the level I assigned. Also, based on my evaluation of the platform, I think Del images and lexico dot art. Are more beginner friendly because they have simple interface and just easy to use. Then for dream studio blue willow and mid journey, it's more intermediate, especially for blue willow and mid journey, you need to create a discord account, Leonardo AI. I would put it somewhere between intermediate and advanced because I think it has too many features for a beginner user. For Automatic 11, 11, that's definitely for advanced users. So let's check out some pros and cons for each platform. Okay, here are the pros and here are the codes for Ali. It has simple interface, it's easy to use, but the images are pretty low quality and the plan is quite expensive. Images right now it's free to use. It also has simple interface. Some limitations here is that there are no profile with history of generations. Also, it's sensitive to some basic words, some things you won't be able to generate with images. Lexica, it has simple interface, it generates high quality images. It is also an image search engine where you can look for image and prompt inspirations. It's great with short prompt. It allows to generate images in a private mode with a paid plan. The basic plan is pretty cheap. The limitation here is that it only uses its Lexica models, which give a specific aesthetic to the images, which may not be great for all the art. The next platform is Dream Studio. It was created by a company behind Table Diffusion. Here you get access to newest table diffusion models. It also has some advanced settings for S. It has difficult UI and also limited models then Blue Willow. It's beginner friendly and it produces good quality images. But some limitations here is that it's on discord, so you need to have a discord account, and it also has limited settings. There is little control over image generation for mid journey. Mid journey creates really high quality, realistic images. It's great with short prompts. You can generate images in a private mode and the basic plan is also pretty cheap. Some limitations here is that it's also on discord. You need to have Discord account. Right now, they don't have image editing yet except for Out painting. If you need to change small things on the images that you generate with mid journey, then you would have to use a different platform. Our next platform is Astra. It has simple interface and it's best for high quality custom model training. It's also based on stable diffusion, it's also pretty cheap for the cons here is that the settings it has may be quite limited. For a more advanced user. Leonardo I has many settings. It has a lot of stable diffusion models. It allows you to generate images in a private mode, and the basic plan is also very cheap for the cons Ganado is still in the wait list mode. If you want to use it, then you should register in advance. Also, it has quite advanced. A user interface has a lot of features and settings, so it can be quite overwhelming for a beginner. Then our last platform here is Automatic 11 11. It has high image customization and control, lots of settings, features extensions. A big benefit is also that you can use your custom models for the coins. Automatic 11 11 is not beginner friendly, it's definitely for more advanced users. In the next few slides, I will show you what images I was able to generate with most of the platforms here. Okay, here. The first one is L here. As you can see, I've tried to use different art types to showcase the platform. Okay, let's move on to the next one, Lexica. Especially with the portrait, it has this really nice aesthetic Blow willow, here are some images. Then mid journey here we really get creative with some of the artwork. Adobe Firefly. I found that it's really good for logos and illustrations, but right now it's not for commercial use yet. Then there is Leonardo. The images that I generated here use the Leonardo diffusion with prompt magic. Then here we have automatic 11, 11. Every image I've created with different model. Here are the settings that I've used and the models. Now I want to show you some image to image generation with different platforms. Here is the original image, Here is the result with different platforms. This is Al here. It wasn't graded with proportions at all, The body figure is all distorted. But Lexica and mid journey were really good with more simple image here. Let's see, the results here is with day to lexica and my prompt was iridescent wolf magical colors. This is the result with Lexica mid journey and Leonardo with Daly. You cannot actually add your prompt here. It just creates variation of the image. 3. AI Image Editing Apps Introduction: Now let's move on to photo editing platforms. Before we've discussed the AI generating platforms, here are photo editing. Actually all of the photo editing platforms here also have image generation. As you can see, text to image model, that's the image generation. The clip drop uses stable diffusion until the Firefly has its own proprietary model. Runway is stable diffusion based how easy it is to use. They are all pretty simple to use. Now let's take a look. What features do the use platforms have for image to image? Here we have only runway for image variations, we have clip drop here in the brackets is the name of the tool here. Reimagine Excel allows you to do image variations in clip drop in runway. It's image variation for out painting. Again, we have clip drop with the tool called crop in runway. There are two tools that allow you to do out painting, extend image and infinite image. For in painting, we have adobe firefly and runway. Here with firefly it's called generative fill in runway it's called rays and replace. Then we have clean up. The difference between clean up and in painting. In painting allows you to replace the area that you don't like with something else. Whereas clean up will just erase the thing that you don't want to see on the image. With a clip drop, it's called clean up with Adobe Firefly. You can do that in generative fill. Replace background, you can do it in clip drop. The tool is called Replace Background in Adobe Firefly, it's called Generator Fill in Runway, it's called Backdrop Premix. Or you can also use a different tool called Creator.com to remove background in your image, you can use Clip Drop. And the tool is called Just Remove Background, or you can use the website called Segment, Anything.com by meta. The next feature is Up scalar. For Clip drop, it's called Image Scalar Runway, Upscale image, or you can also use a different website called piggpgt com. Here are some other tools that each platform has here. For clip drop, we have relight and text remover. For Adobe Firefight, we can generate cool text effects here. They also have generative recolor that allows you to change the colors of a vector image. Then for runway, we also have a color frame interpolation, three D texture, and model training. Another website that I also want to talk about is the vectorized I, that allows you to create a vector image from GPG PNG image. In the next few slides, I will show you how some of these features work. Let's take a look. Photo scalar, Here is my original image with a low resolution. I wanted to upscale this photo. Here are the different platforms that allowed me to upscale the photo. Here are the different results. My favorite one was the clip drop for X, smooth for artwork up scalar, this is the image and here again, different platforms, this time for artwork upscaling. I like the big GPG way more than the other ones. For background replacement, here is my original image from mid journey. Here are some images that I could get with different platforms. Here I use the Create It. It produced nice results, but in lower solution. Here is the clip drop. Here is Adobe Firefly and Runway. As you can see, Runway tried to add the platform here to my product. Next, let's move to Image variations. Here is my original image. I generated it with my journey. So let's take a look at what results did we get from different platforms. Okay, here we have clip drop with the re, imagined Excel tool and Runway. Here we've got in very similar style but now the furniture is arranged in different manner. We're asked for runaway. We've got the same composition but just slight modification in color and style of the furniture for out painting. Here is the original image, you are probably familiar with this meme. Let's see how different platforms extended the image. Here we have clip drop with the tool called and crop and runway with the tool called Extend Image. Here are different results. From my experience with clip drop, you would get a bit better out painted images than with runaway with a fewer artifacts. But still is not great especially keep. You see the hands and the legs. Next is remove background. Here is the original image here. I chose the image with the hair because it's the most difficult thing to remove. Background with hair here, let's see, here is the result with segment, Anything.com and here with Clip drop. Remove background here, it's apparent that Clip drop did a way better job. Now let's remove background from a more simpler image. This is the original image. Here is segment anything.com and clip Drop. As you can see, the clip drop doesn't have these pixilated areas at the bottom here. Again, Clip performed better here. Here I've compiled a list of all the features that you will learn in the scores so that you know what platform you can use for specific feature such as remove text, You can use drop here, let's see. For example, colorized black and white images. Then you can use Runway Had Color. For some, there will be multiple of platforms that have that tool. For example, train your model. You can use Runway, Leonardo or Austria on that node. See you in the first module. 4. DALL-E Introduction: Hi everyone. Today I want to start off by introducing you to Dali, which is one of the simplest AI image generation platforms. Dali stands for the noising order encoder for learned language embeddings. It was developed by Open AI. It's the same company behind Chip. It was introduced in January 2021. It uses an algorithm similar to stable diffusion because Dali relies on a process called diffusion. The image generation starts with a random set of noise, which is basically a random arrangement of pixel values. Then this gradually is modified in a series of steps to make it match a given prompt. By starting with a different set of random noise. Each time different images, different results can be created from the same prompt. The process is called diffusion because it involves spreading out changes across the image. Each step of the diffusion process makes small adjustments to the pattern, making it more and more like the desired image. You can think of it like looking up in a cloud sky and finding a clown that resembles some object or an animal. You can make it more clear and defined in your imagination. That's basically the diffusion process. More advanced model became available in April 2022. That's what is being used right now, it's Ali version two. Now I would like to outline some advantages and disadvantages of using the platform to start off. One of the advantages is that it's very easy to use. If we go to Ali, you just need to type your prompt here and click generally. There's no extra features. Very simple to use for a beginner that it will generate an image according to your prompt. Another good thing is that it gives some free credits every month. You can explore the platform for free. Also, it can generate images in different styles. If we go to the gallery, you can see images in different styles. I would say realistic styles like this one paintings realistic impressionism, cartoons as of styles, which is great. Another great thing about Delhi is that I can upload and edit images. If I go to the website, there is a button to upload an image. I can upload an image here, I can make variations of my image or edit the image here, there are two ways I can modify my image. First is in painting, basically I can replace certain part of my image with something else. For that, I just need to erase the part of the image that I don't like. Write a prompt, it will generate a different image with this part. Another feature is that I can make an out painting, basically extend the boundaries of my image. Again, I can write a prompt and it will extend my image according to the prompt that I write. Explore in the later videos. In terms of limitations day produces quite simplistic images compared to other platforms. It doesn't have as much details, it's not as creative. Another thing, it only produces square images if we go to the gallery. As you can see, all the images are squares, for example, if you want to create a landscape dimension or a certain aspect ratio. Unfortunately, here you cannot do that. It's only going to be a square format. Another problem with Dell is that it cannot quite produce potter realistic images, because when it renders a phase or hands, it makes mistakes. Let me show you what I mean. This is a problem I tried and this is the image I got. If we zoom in, you can see some artifacts. The teeth are not friended correctly. The eyes I'm not sure if it's drawing, so there's big problems with her facial features. Yeah, I wouldn't recommend using deal for high quality, for realistic images. Another thing with Ali is that similar to stable diffusion, it requires detailed prompt. If you want to create an image with more details or you are looking for specific image, then you need to write a very clear detailed prompt. Otherwise, you would get very ambiguous results. I want to show you these four images were all created with the same prompt, a girl in a scarf, and as you can see, the results are all over the place. We have some realistic images, we have cartoon. Make sure you add as much details as possible and open actually recommends to have more detailed prompt. Even if you write a very detailed prompt, the application sometimes may struggle to produce the desired output. Just because it may interpret certain things differently. And you may need to iterate a few times and generate more images to get what you want. This is very similar to other image generating models. In the next video, we'll start exploring Ali and I will show all the features in detail. See you then. 5. DALL-E Image Generation: In this video, let's dive in into Dali and let's start exploring it. The first thing you'll need to do is to go to the website and this is what you'll see. You can read more information about Dali. Here's an introductory video and some articles about Ali and latest updates. It also outlined some features such as out painting in painting. This is what I covered in the first video, variations of course, image generation. What I want to say is that when Daly first came out, it really amazed people. Because the way it combined different objects, concept attributes and styles in a surreal image, and yet making it auto realistic, that was only possible in our own imagination, in our head. And here we go, we have the stool that can take our imagination and put it in a canvas. So what Daly is best for is making this crazy, unreal, surreal images. You can put lots of concepts into it and make your vision come true without further ado. Let's go ahead and login or sign up for Ali. If you already have an open AA account, for example, if you use GPT, you can log in with the same account, otherwise you'll need to sign up. But it's very easy, All you need to do is give your e mail address and they will also ask for your phone number. I will log in after you log in or sign up. This is the page you'll see. Let me just talk a little bit about how it works here. This is where you will write your prompt. There is also a feature to surprise me that will just give you a random prompt. If you need some inspiration, you can click on it and it will give different prompts every time. For example, a submarine, a bowl of soup, that's also a portal to another dimension, digital art, for example. This sounds really interesting to me. You can go ahead and click Generate. After loading, it's going to give me a few images. Okay, here we go. We guide our Bowl of soup. That's a portal into another dimension. Let's say if you like a certain image, I like this one for example. I can click and I have an option to download it. Or we can make variations by clicking. Variations is going to make a few more similar images to the one I chose. Okay. As you can see, the style is very similar to this image that I chose. It's all black background. The ball first. A little bit. Yeah. As you can see, it's not as detailed as it could be. Okay, now if we go back, I want to talk a little bit more about what styles we can create with Ally. For this, it's best to use the gallery as inspiration because it has really good examples and actually really good examples. Because I would say 60, 70% you get really strange outcomes. For example, this one is a three D render. This one is an expressive oil painting. This one is a photo, it doesn't say the style. This one is style paper wave. This one bang style. As you can see, every image is produced in different style. The style is not specified. Sometimes it's the artist who is specified. Johannes Mu is the artist of a girl with a pearl. As you can see, the style is very similar, but now it's an animal. Again, here's a hand drawn sketch. This one is a photo cyberpunk. This one is an oil painting. An oil pastel. This one is a cartoon. This one is a pencil. And what a color drawing again, This one is a three D render. This one is a comic book style, let's say a certain style of the image, for example, the style. You want to generate images with the same prompt, then you can go ahead and click Try this example. This will images with the same prompt. If you want to modify the prompt a little bit, you will need to copy it. Let's copy it and base it here. Let's say you want a cat with a pearl earring and click Generate. This is what we've got. These images are AI's interpretation of a cat with a pearl earring by Johannes Mir. As you can see, all these images are quite similar and yet different. On every image we see a cat wearing some blue scarf and pearls. Well, this particular image does have pearls, but overall the style is very similar to Johannes Mir. This is what the Dali algorithm does. It takes the prompt, interprets it, and then gives off different images. Now we can try some more examples and try to challenge AI to see what prompts it will give. Let's try a giant bearing, riding a bicycle and a cartoon style generate. Okay, great, I actually like the first image and the last, although it still needs some editing because you can see the bicycle handles are not rendered well, and the E as well, as well as the last image, eyes are not rendered correctly. It needs some editing. The others, I would say, are worse just because they like the details. As you can see, again, the cartoon styles are all different here. If you want a specific style, then you need to specify it in the prompt here. Let's try some more examples. Let's do, for example, animals Atlantis lost city of Atlantis. Let's make it a digital art. Okay, this is what we've got. On the first image, I actually see whales here. This is probably some architecture. This is all underwater. I got the concept of City of Atlantis correctly. However, all the other images, the second, third, and fourth, they are pretty bad if none of those looks like animals. N here actually this is something that you'll experience when you use AI generating platforms. Because some images will reflect your concept or your prompt really well. However, you will also get images that are really wrong. This is just a heads up, it's going to happen, you'll just have to generate a few more images, because the more images you generate, the higher the likelihood that one of these images will capture your concept correctly. Actually, in Ali, if you don't like any of the images, you can go ahead and click on it. And there is a flag button. When you click on the flag button, there's two options. You can flag the image if it doesn't match your text description, which it doesn't. So I can go ahead and click it and hopefully open AI. Team will go and review the images that helps the team to build a better product and improve their algorithm. That's it for this video. In the next video, I'll go more in depth into how to make editing with Ali. 6. DALL-E Image Editing: In the last video, we tried a few products with Dali. And this is where we left off in this video. I want to show you more of Dali features. Let's go ahead and try all the features that Ali can offer to us. I want to make some edit on the image that I like. All you need is to click on it. It's going to expand it. Here's a button called Edit. We can click on it. As you can see here, the interface is very simple. There are only five buttons. And in order to edit the image, we need to select an eraser. So make sure eraser is selected. And when you move to the image, you'll see a wide circle and you can start erasing things that you want to replace in this image. For example, let's say if you made a mistake, you can always go back by pressing control on Windows or command on Mac on the right hand panel. We can also change the size of the eraser. For example, I want to make it a little bit smaller to make raising more precise. What I want to remove from this image are those squiggly lines. I'm not sure what they are supposed to be, but I don't like them in this image. I will remove them, and when you remove something, prompt will pop up and this is where you will write a prompt to fill those ******. I'll put underwater fish for example, and click Generate. After rendering is complete, you will see four different images that fill out those ****** in a different way. If we zoom in, these are the ****** I didn't like as you can see here. I fix those ****** and added a fish, I believe here. Yeah, it added a few fishes on the other image here. Now we can go and select the best one. I like this one the most. I'll go ahead and edit this one again. Now overall I like with the design of this image, what I would like to do is to expand the image to show maybe some more information. To do that there tool to make out painting and that's called a Generation Frame. Click on this button, you will see a square like this. You can drag it anywhere you want and place it anywhere you like. You can also zoom out the canvas a little bit. You have more space to create. When you zoom out, there is also this button pen. This allows you to move the canvas anywhere you want. Let's move it here. Now let's click on Select. This will allow us to select the frame again and place it anyway we want. Let's start with the frame here. I want to expand my image to the right. And let's zoom a little bit, maybe not that much. Okay, now let's write a prompt for this new generation frame. I want some more of this image. I will put Underwater City. Let's see what it's going to generate. Okay, interesting ideas here. As you can see the extension. This image is in the same style as the original image. Now again, it has four variations and we can see which one we like the most, like this natural curve here. After you look through all of them, choose the one that you think suits your image the most. And then click. Except if none of the images look good. Then cancel and try again. I'll click now. Let's do the same to the left side, move the canvas alto bit. Let's select the generation frame and move it to the left. Another thing what I want to say is it's important to capture a little bit of your original image. Because that's how AI knows in what style to create this new image. If you do like this, it will have little information I would always recommend to do at a third, to capture at least the third in your new generation frame. For example, let's do this, let's capture, let's put under water, lost city of Lantus. See what it's going to generate. Actually, this is better than what I expected. It still captured the style because we had a little bit, lots of cool ideas here. I like all of them, but I would leave this style very interesting idea here. I'll accept it. We're very likely that this is what it generated because it took very little information about our original image and created actually very similar style. All the details actually look very similar to what we have here. However, if the generation frame doesn't touch your actual image, then it will generate things in a completely different style. Let me show you. Let's zoom out. Let's put generation frame. Let's put generation frame completely unrelated. Let's do the same underwater Los City of Lantus and click Generate. As you see, let me as you can see this image is completely unrelated to our style, totally different. It's just a brand new AI's interpretation of underwater Los City of Atlanta. If you want to expand your image, make sure the generation frame overlaps with the image. Let's cancel this. You can go and expand your image up or down. But this image, I think we did a pretty good job. Of course, we can do further editing of some fish. Let me quickly do it. When you editing, make sure you return the frame back. So if you don't have your frame at the places where you erase things, it's not going to generate image here. It will generate image wherever you have the frame. I'll move it here first. Okay, This is something I like accept. And let's move the generation frame here. Okay, after a little bit of editing, I think we have a very nice image here, so we can go ahead and download it. So you can click on this button and save the image. This is what we have After all the work and editing, I think this image came out very well. That's how I would imagine the lost city of Atlantis. 7. DALL-E Outpainting: Now let's try something different. Let's applaud an image, for example. This one deal asks if you want to crop an image. This is important if you want to make variations. Because Delhi's limitation, it can only generate square images as an input, it also square images. If you don't want to make variations, then you can skip cropping and basically added the image the way it is. However, if you do want to make variations, then just what you would like and click Crop and then choose Generate Variations. Okay, This is what we've got. As you can see, the images are really bad because Ali is not meant for photorealistic styles. But let's try something different. Let's uploading an image of a famous painting of a starry night, for example. Let's crop it and generate variations compared to the ballerina. It did a way better job with this, with this style, just because it's more abstract. It looks really good here. Okay, now what I also want to show you is that you can take this image and if I go to the editing mode here, I can actually upload more images. Let's first zoom out a little bit here. I can upload some more famous art. I can upload an image like this. As you can see, it's way bigger than this one. I will resize it to match the other one. I think this looks pretty good. When you're done, click the place here, check Mark. Now let's zoom a little bit. And let's upload one more image. For example this one click place. Now I will go to the generation frame and put frame where both images overlap. For example, here I'll put an impressionist painting of, of it. Try to merge them. Quite interestingly, I like this one. I will quick accept and I will remove the sport. I'll go to razor and I'll remove the sport. I will move my frame down and repeat. I'll put impression painting of a down at night. Okay. It actually match pretty well here. Let's see what are the other ones, okay? I think this one is the best one. Let's accept. Let's accept again, let's erase the part. Let's continue like this. I think this is the best because it has street going up where you can see the town. I'll set the win in order to merge this painting. It would be best to erase some sharp ends just because it, um, separate everything. I'll just remove the sharp ends here. I will go and put my generation frame here. Presion painting of a town at night. Let's see what it's going to generate. As you can see, it detected that these two styles are very different. It made a clear separation here, but on the next one, it's pretty well done that it detected that this is a sea or an ocean. You have a little bit of a beach here. My favorite one is the second one. I'll accept it. Let's just continue and see what it'll come up with. I think this matches the style. Let's do this one. Let's turn a little bit more here, okay. Wow, that's interesting. Except some boats here. It's either number three or number one here. Definitely way more people. I think this one is more clear. Accept this one. This one, I'm not sure if it's a car or something, but why did it match it like this, kill it's cancel it. Okay. Understand. Because the part is very sharp. In order to help I merge it, let's erase the spart. Let's delete the sharp here. It will give AI more ways to merge the two images. Let's generate again, maybe a person will be noisy here, except let's do our final one again. Let's raise the sharp edges here. Let's again select, I'll move the frame a little bit here. Let's generate again. I think this one is the best. Yeah, except as you can see, Let me move the generation frame over there. As you can see here, we took three different paintings and combine them together in one image. I think this is fascinating because two of the images were Bango style. Another one is a painting of Monet. It is quite different, but somehow AI was able to join everything together and create, in my opinion, a masterpiece. In this video, I used images that were quite similar in a style. However, you can try something different. You can try completely different styles, completely different themes, and combine them together. And that's what Out Painting is for. It's now up to AI to think how to merge the two completely different things together. So here you go, Have fun with it and see you in the next video. 8. Prompt Examples: To conclude Dally's module, I would like to use different prompts to show what Ally is capable of. I chose realistic photo, logo, magical realism illustration, landscape and conceptual art. I think that is more or less a good representation or various or genres because our course covers a lot of AA or generating platforms. Each platform or model have its own strength and weaknesses. So the best way to show you the differences between the platforms and where each platform performs best is to use same prompts for all of them. So this is what we'll do, starting from Dali. Let's get going. Let's start with realistic photo. For this, I choose a portrait photograph of a young woman. So I wrote professional portrait photograph of a young British woman in a jacket with wavy blonde hair, beautiful symmetrical face. Cute natural make up, blurry, raining, city, street background, highly detailed Sharp Focus Depot field, and then this is aperture and the ****. In the next module, I will explain why I chose certain words and how I wrote this prompt, but for now, I just want to show how Daly will interpret this prompt. Let's go ahead and try this prompt with Daly. Let's paste our prompt and clear Generate. All right. On the first image, I think Dali did pretty well compared to everything else. Here, it has the most natural look. However, there are still some problems, some inconsistencies with eyes, maybe lips as well, but as you can see, because the prompt is longer and here I write that it's highly detailed. It would usually give better results compared to a shorter prompt for portrait. As you can see, Daly is not good at producing photo realistic images of people. It makes a lot of errors. Okay, let's strike a different prompt. Let's do a logo here. I wanted to make a logo for a bakery. I wrote line logo for a cupcake with a cherry on top, clean lines, simple shape, minimalist vector. Let's try this prompt with Daly. As you can see here, it did exactly what we told it to do. It's a line logo. Here I can see clear line, simple shape. I like the third one the most. Also you can see the name, I don't know, it came up with the name for my bakery. I like the font here as well. Maybe I would remove the line, this line. But overall, it did whatever we wrote in the prompt and it was creative here. It doesn't look like a cupcake, but it has a cherry on top. It looks more, I don't know, like a burger with the cherry on top here, it added those lines. I think it did a pretty good job with the logo. It generated a few interesting ideas here. Let's try something different. Let's try magical realism. Magical realism is when you take something real, for example, like a dog and place it in unreal situation. For example, dog riding a bike. In this case, I decided to do three D Render or raccoon reading a book, armchair lighting from a lamp, realistic unreal engine. As you can see here, deli captured exactly what we wanted it to. I wanted a lamp con, reading a book armchair. It perfectly did the job here. However, when we zoom in, the eyes are missing here and you can see certain artifacts here and strange lines overall. It's not great because it lacks details, it's poorly rendered, but conceptually is okay. I found that if I remove this from the prompt, the lighting from a lamp, it generates better images. I think this is a little bit better. We have a raccoon in glasses reading a book. There's a little bit more attention to details, although still it's not too smooth. But compared to this one where there are no eyes at all, I would say the more simpler one is better. And one of the reasons is that here you can see raccoon is much closer to us. Here it's further away. The closer the object, the better it's rendered, the more details it will have. Just because of the way model works overall, Deli may be a good option for simple three D renders like here, but it would still require some further editing and retouching. Let's try something different. Let's try illustration. For this one, I chose children's book illustration of a girl riding a bike in summer. And here are the names of illustrators Axel Scheffler and Non Blake. Let's try that one. I think these are quite good illustrations. If we zoom in, I don't see any problems. This is something I would see in a book illustration, for example. Very simple. It follows exactly what we wrote. It's a girl riding a bike. I like those three here. This one I think the girl is missing, and nose. With some editing, I think it would still be a good image, although we would need to fix the leg here as well. But the this one and the third one, pretty legit and could be used right away for book illustration. I think Delhi did a fantastic job here. Let's do landscape for landscape, it's digital art of magnificent medieval castle between the hills and fields. Large pornographic background with dense nature and mountains, grand fortress, epic scene fantasy. Let's try it. Oh, wow. I'm very impressed with the first image here. I think this one stands out the most just because of the lighting. See how it's dark here and the light falls on the grass, this area here. And I think this makes it magical for some reason. We can see here, the image is in different styles. This looks like a oil painting. I'm not sure about this one. This one looks like pastels. This oil might be acrylic style, but as you can see, these use different brush strokes. I want to mention most AI models are better with landscapes compared to photo realistic features like photographs of people. Our face. There are certain features like eyes, they have to be the correct proportion or hands. We only have five fingers, it's not six or seven, I Sometimes we will get those things wrong, makes mistakes. However, those mistakes you wouldn't see on landscapes because, for example, the shape of the cloud is not as important as the shape of an eye. For example, if it makes a mistake, that mistake would actually look like a creative interpretation or creative element of the artwork. Mistakes here would look aesthetically pleasing compared to the postic phase features where we would immediately spot the mistake. For this prompt, Landscapes Tally performed really well. In the next prompt, we have conceptual art. Here I wanted to challenge AI. I didn't include any subject, I just include an idea, the meaning of life. And I wanted to see how AI would interpret it and put it into art. Then I just put adjectives. The whole prompt is the meaning of life, Breathtaking art. Standing high resolution, highly detailed, inspirational Eight K. Let's copy paste our prop here. Oh, wow. I was very creative with these prompts. As you can see, all these four images are in completely different or genres. I absolutely love this one. Here we have emerged of a person's face with a sky. So I would interpret it as a God figure. I'm not sure how AI even got here, but I think this has a very deep meaning. This one nurtured. This is a planet, maybe looks very futuristic. This one is a landscape. Some shapes, abstract art, fantastic. For this prompt, I'm truly amazed how creative A I was when I played with AI before. For the same prompt, it generated this wave with light, which is absolutely beautiful. Also, this brain with neurons. It's astonishing how AI takes this prompt, this idea, the meaning of life, and makes an art out of it. Because it's usually us who were taking an idea and making it into art. But now here we have AI that creates crazy cool things. And now we can look at those images for hours and try to interpret them. So for example here, the color scheme is so interesting. Now you know how to use deli and what art it's good for, it's strength and limitations. For example, I would use it for simple three D rendering. I would use it for cartoon style images. And I would use it for landscapes because I think it does a pretty good job on them. However, I wouldn't use any portraits not photo realistic. At least because they're way better platforms that can do photo realistic images, portraits of people. On that note, I will leave you with Ali and I will see you in the next module, where we will explore prompts and actually how to write them. See the. 9. New Update: Comparison Between DALL-E 2 and DALL-E 3: Hello everyone. This is an updated module in this presentation, the images you see here, they were all generated by a Gi. But this is not mid journey, this is not, Leonado, guess what? This is Daly three. And as you can see, it has improved so much from Daly two. Okay, so let's check out the differences between dally two and Dally three. Okay, with my presentation on the right, you will see images generated with Dally three. For your reference, I also included prompt. For example, here we have a prompt, Make a logo for a coffee shop with a name Espresso Club. Let's get into Dally two versus Dally three. Well, the resolution for images is way higher, Daly three generates images with the resolution of thousand 24 by thousand 24 for a square image or for landscape in portrait, it's 1792 by thousand 24 pixels. The resolution is twice as higher than it was in two. For al two it was 512 512. So you see a significant pump in resolution. Okay, next one is the improved details. Because the resolution improved, now we get more intricate details in the images. Next is the superior image quality. Dali three, I believe, had a way more training, training images and the images, it can now generate a way better quality. Especially, it's noticeable for the portraits because when I used to, there were many artifacts with the faces. Now it's not a problem. One of the big updates is the legible text, as you can see on this image, as I've asked it to be with a name Espresso club, that's how it appeared. Espresso Club. Of course, sometimes you do get mistakes, especially it sometimes would duplicate the letters that I notice. That's the most common mistake, but then you get lucky and get the name as you've requested. Finally, it can accurately depict historical figures. Let's see, here's the prompt that I gave to Daly. Three, make an image. How would a girl from the famous A girl with pearl earring by Johannes me would look like? Now here on the left we have the original painting by Johannes Mia. On the right we have the image generated by Daly. Three definitely knows how a girl with pearl earring the original painting looks like and it made a replica but this time it put in the scenario where I wanted to be in the modern times. It added those elements. As you can see, it added the Genes Jeans jacket with the metallic pattern here. If you look closely, look at the texture of the jacket, texture of the skin, how the light falls on the skin, especially like the touch for the lips, Also the texture of the scarf. Here, it resembles the scarf on the original painting. This is something that we would never get with Daly. For that, I wanted to draw the parallel between Daly two and Daly three. Let's get into that. Here on the left we have an image that we generated with Dally two and on the right we have the image for the same prompt generated by D three. The prompt was professional portrait photograph for a young British woman for Dally Tube when we were generating images. Actually here I chose the best one because most of the images had artifacts, a lot of errors for facial features. Even here we do see that eyes are not symmetrical. That's. Artifact by Daly two, whereas for Dali three, now I do not see any problems. And that's actually the first image that I got from this prompt. We also have more details. You can see the zipper on the jacket. Look at the texture of the sweater even though the background is blurry. But you definitely see details of the buildings around buses, cars, and just the street. Just a lot of more things going on for our next prompt, children's book illustration for a girl riding a bike, the same thing for Dally Three, We get more details again for Daly two, this is also one of the best images I could get for my prompt. Others had problems and artifacts like with the legs or with the bike, maybe nose, mouth, missing eyes and so on, but with Dally three, those artifacts are more rare on this image. We actually, I think we're missing girls mouth but it's not that noticeable here. Again, we have way more details, but at the same time it used this pastel color children's book style illustration. Our next prompt was Magical realism. Read, render, or Racon reading a book, arm chair, lighting from a lamp realistic and real engine. Here, again with Ali to we have some artifacts. If you look at the eyes, the ears, maybe the arm chair here, definitely a lot of room for improvement. Whereas for Dally Three, we get a completely different image. Look at the texture of the raccoons fur. It's so realistic as well as the lighting. We have this lamp as our lighting source and everything seems proportional and correct in terms of composition. The next prompt is line logo of Cpk with a chair on top, clean lines, simple shape, minimalist vector. I pretty like the results we've got with Ally two. But of course Daly three is even better because here you can actually see that the name of your company will be Legible Cupcake here. Another great thing is that sometimes when platforms are advanced and then they can create advanced images, they try to add a lot of details. Here, I ask for a line logo of Cupcake. I want the image to be a line logo. I don't want it would be a three D cupcake or a super complicated logo. I just want very simple and what I liked about Daly Three that it actually followed my prompt and gave a line logo. The next one is the landscape, We have digital art, magnificent medieval castle. On the left we have a rough sketch of the castle which has a place to be, but on the right image by Daly Three, we have a lot of details on the castle. Look at all those windows, the houses, we can even see the windows on those houses. Again, the composition is correct on the background, the objects are light on the foreground, the objects are more saturated with Daly Tu. We could only make square images, whereas with Dali three Chagpt we are able to make a landscape size image which is great for landscapes. The final prompt was, the meaning of life, breathtaking art, standing high resolution, highly detailed, inspirational eight. I liked the image that Dali to generated. Looks very interesting with neurons here, brains, and so on. But with Dali three, the image was exceptional because phoenix is a symbol of being reborn. And also the colors of the image are fantasy like pretty cool. Dali Three was released in October 2023 and it's available in being as image creator and part of being chat. It's also integrated in Cha PT and Enterprise. What are some limitations of Dali three? Well, if you want to use Dali as part of a GPT, then you need to get a subscription and upgrade to Cha GPT. If you just want to try out Dali three, you can go to being AI and try it for free. The next limitation is time to generate. I noticed that with Cha GPT it could take around 30/42 to generate, which is pretty long for only one image. With being is a bit different, it generates four images at once. I would also say it takes around 30 seconds. But there are boosts that allow to generate images a bit faster and we'll talk about that later. Another limitation are the mistakes it makes in the text. Here you can see I asked it to make a poster of Hubburn. The nice thing, it knows historical figure, it knew the deryburn. But the problem is the text here, if you look far away, you would probably see something written like Dre burn. But if you look closely, you definitely see mistakes. There is double D, double double B U R. Again, the most common mistake is that it repeats the letter if you get the name would be written correctly for image editing right now, it's only possible in GPT and how it's done. First you write a problem to generate first image, then you write a follow up prompt. For example, saying here I can say, please improve the text and hopefully it will generate the same image but with improved text. But here is the limitation because for the follow up prompt, you may get the same image but with the edit, or you can get a completely different image. Because there are no settings, there is very little control of what's done. Your best bet is to describe the way you want the image to be edited as precise as possible and hope for the best. But that sometimes doesn't work, and for that reason, using other platforms for editing may be more easier. Lastly, we have policies. There are certain images that you cannot generate. For example, no explicit content, no copyrighted material, no offensive content. It won't generate images of modern politicians, public figures, or recent artists work. 10. New Update: DALL-E 3 with Bing vs DALL-E 3 with ChatGPT: Now let's try to use Daly three. First, I want to show you Daly three in Bing. Then I will show you Dally three in Chip. The reason for that is that there are different functionalities and I just want to show you how they differ. In here is the link. This is how you can access Dally Three. If you click here, it's going to take you to Microsoft Bing Image Creator. All you need here is to login with your Microsoft account. In the first page, you will see the images that were generated with Ali Three and questions and answers. Let me tell you a little bit more about Ally Three. In being it's free every day, you get boosts that are allowed to generate images a bit faster than usual. You can also exchange your Microsoft rewards for those boosts. Let me show you here, you can see I have 15 boosts. If I go to the questions here, how do Microsoft Rewards work with Image creator here if you run out of boosts, you have the option to use Microsoft Rewards to redeem for additional boosts and enjoy faster processing times when you run out of boosts. An image creator, you'll be reminded that you have the option to redeem Microsoft rewards points for more boosts. To be honest with you, I've never tried to redeem Microsoft rewards points for additional boosts because I generally use Cha GPT with Ali. But if you use Microsoft Rewards, that's a handy feature to know here. What else you need to know about Dali? Three in being here, it generates four images at a time. It only makes square format, there is no portrait or landscape, there are no editing capabilities, and you cannot applaud a reference image. Let's try it out. For example, here we can even start with Surprise Me. It's going to generate a sample prompt for us. Here we have Boho interior design with red accents. Let's try that. Just click Create. As you can see, it used one of the boosts. It should generate an image pretty quickly. Here are the images that we got here. I do not see any popping up mistakes or artifacts unless I look closer and maybe you can spot a few. But the adherence to the prompt is phenomenal. Let me show you a few different pront, For example, I tried a futuristic sneaker, digital art three D render. Let's take a look at those images. Look at the texture of the sneaker and the lights. Definitely futuristic. Let's see the other ones. It even incorporated these dots in the material suggesting that it's a preferable material. As N advertises, The attention to detail here is astonishing. And look at all the three D design it just and if you have a product, you can use Dally three for photoshoot inspiration or for a background, for example. Yeah, here it's very simple. There are no settings. You cannot change anything. You just write a prompt. That's it. On the right here we have history, which is neat. The only thing here you can do is click Save Images Here, we can actually customize that. But that's not the editing, it's just the Microsoft designer which puts the image into a mock up, for example, if you want it framed or you want to make a post about it here. There are some templates that you can use. I'm not too familiar with this because I'm used to canvas. But yeah, this is something you can explore. Now let me show you how you can use being chat to, to generate images. Chat here. Click on Chat. Here, we can, can choose a conversational style. For example, let's use the balanced. Let's generate an image of a cat on a piano. Piano. Why not? Let's click Tab. It will generate an image. The cool thing is that the result of this generation will be seen in this image. Create a platform history. You won't lose. Okay, so here are the results. Let's actually open it in the image creator. Here, let's go to creations. And here is our cat. Let's see, we have the cat piano. This one is pretty cool, not that. Okay, this is how you can use Do three in creator or with the Bing chat. 11. New Update: DALL-E 3 with ChatGPT: Okay, let's now move on to GPT form and Dali three integration. Dali three was natively integrated into GPT GPT enterprise. If you want to use Dali three, you'll need to have a subscription, at least for GPT. What are some features because it's four. I've included some more functionalities that are possible like analyze the images. I just want to show you all the capabilities of this union between P four and Al three with the images. Okay, let's see, generate images from a text prompt. It offers three image size options. 1020, 4,024 pixels square, landscape and portrait. Unfortunately, you cannot use a custom. But at least this is an improvement. Because in Li two, it was only a square. Here. In GPT four, at least there's landscape and portrait. Unlike in being AI, which is also only a square with GPT, you can edit images with a text bond. It may not be ideal, but you can tweak the images a bit and let's say you don't like the colors but you like the composition. Then this is something that you could do with GPT. Make images based on input image. That's also a benefit of using GPT. And Ali, because in being AI, there is no feature like that. You cannot upload an image and ask to generate an image based on your input, whereas in GPT you can. I will show you how also you can analyze input images. This is useful because sometimes I want my image to be analyzed. To know what kind of prompt I can use to make an identical or similar images. Let's explore some of those functions. For example, here I have a reference image. I uploaded an image of myself and I put a prompt, Generate an image of me in a comic book style. This is the result. It produced a comic version myself. I liked how it captured the wavy hair. My green eyes, nose, the shape of eyebrows, if you look at the image, is spot on as well as it captured the black suit. Well, the shirt design is a bit different. Also the background, the park, we captured all those small nuances which is pretty good from one image. In terms of replicating my face in the images, it didn't quite work. I tried some realistic examples. For example, again, I've uploaded the same image. Let me show you, here is the image I've uploaded and I asked, generate an image of me in the comic book style. Here is the result. Then I asked it generate an image of me, but if I were in 18th century, and here's the result again, It captured that the hair is curly, the eyebrows, but just the facial pictures do not resemble me. I guess it would be great. If you want to make yourself into a fictional character, like a comic character or a cartoon, then it would work better than the realistic style. Well, at least for now. And then I also tried more images. Here's the result. Again, no close to facial features. This is how you can upload your own image and just try out and create a Comic books of yourself, for example. Another reference image I gave was this tower in Estonia, In Talin, I gave the image and I wrote the prompt. Make an image similar to this one. Here's the result. I love how it captured the tower. Also, you can see on this the original image, this building with the spike. You can see that it's exactly the same one with the spike here, which is great. Let's try with a different image, applaud an image, and see what PT with Ally will generate. Okay, let's open a new chat here. You can attach a file here. I have a different image I took in Estonia here. It's going to upload the file. While it's uploading, let's write a prompt. First of all, I want to know if John PT knows where I was. Let's ask where is this place? This is the analysis of the image. First. We actually got an error here saying that it couldn't open the Hague format. Let's change our image to GPG. Okay. I converted the image into P, G. Let's use where is this place? Okay. Okay. This is the response, I'm not sure here showing the error, but the response is good. It says the image you've provided appears to be to show the viral gates which is part of the fortifications of the old town of Tin, the capital city of Estonia, which is correct. Okay. It completed the analysis part correctly. Now, let's ask to generate an image similar to this one. An image similar to this one. Of course, you can make some modifications. For example, you can ask, generate an image similar to this one, but during the nighttime or during the spring. Let's, But somewhere let's. Okay. Here we've got the image. Let's just try to compare. Here we have two towers. Well, they look on this image bit bigger than here, maybe that's the perspective. We have people market overall. Again, it generated a very similar image. Which is great because now we can give any image as reference and ask to generate or use elements of our reference image for example. Now let's move on to editing. The image I want to edit is this landscape. The prompt is a familiar one digital art of magnificent medieval castle between the hills and fields. Large panoramic background whose dense nature and mountains, grand fortress, epic scene fantasy. I want it in landscape size 1,792 by 1,024 pixels. I started with the prompt. This is the image I got in the chat. The next thing I felt missing in this image are more rose colors, like purplish, maybe like a sunset. That's exactly what I put in my chat. I asked it to make it with more rose colors. This is the result that Chan GPT gives here the castle looks quite different. In the previous image, it was in the way the fortress was round, whereas this one is more square. As well as we do have different elements such as this additional castle at the top here, but the composition is overall the same and this very similar angle, we got our rose colors. Purple, predominantly purple, pink. Okay, now I wanted a more zoomed in image of the castle. I just As this zoom into the castle. Here's the result I got. I wasn't satisfied with this at all because the image is completely different, even though it says here that here's the zoomed in view of medieval castle, focusing on its intricate details and the rose hues. Here we have monotone brown image. If you are unsatisfied with the result, you have a few options. First, you can regenerate the response, and sometimes that will give you the desired result on the second or third try, or you can change the prompt. Here we have a pretty short. If I want to keep the same image as here, I would put more precise description such as saying, keep the same image, but zoom into the castle. Something like that. Let's try some more editing. Make the castle, make a mistake here. Make the castle more magical and with beautiful nature. Here is the result. We go pretty magical colors. I like this image, but now I want something a little bit more realistic. I make it more realistic while keeping fantasy like elements. Here is the result which I liked, and here we still have magical colors. With the purple blue, we have a beautiful pink color of the castle, but it just looks more realistic as well as like this game with a little light. Here we have blue and in the front we have pink. Okay, let's move on then. I asked to make it look like Disney Castle. Make it look like Disney Castle. Here is the result. This is how you can write simple prompts to go and develop your image. Which I think is great because you start with one image and then through some prompts, you end up with completely different results. And maybe that's the direction you never thought of going. But this is where it took. You love the result, which I think is a way better process than just think of a huge prompt at the beginning. For me, especially simplifies the process of creating a way. You start with something short, like a castle, and then you add more details or you decide which direction you want to go. The next feature that we've already tried is the analyze here. I gave it a strange image, it gave me a description, and I just put that description into Dali three. This is the result it gave, which I think is pretty similar results. We still have the gage, we have the track, the background is different, but overall the feeling of the image is the same. Let me show you how you can do something like that with GPT. Again, let's create a new chat here. I will upload an interesting image and let it analyze the image. I'll make a description for the image. Okay, the text is pretty long so what I'm going to ask is make a description that I can use for a prompt for AI or generation. Okay, we still ended up with a huge text here. What I'm going to do is I'm just going to limit it to one. I'm going to just change it here. I'm going to put limit to to 30 words. Okay, So this is something we can work with. A ring master in a red jacket and black pants, tipping his top hat while aligned playfully bites his head on a circus pedestal with a blurred audience behind. Perfect to the point. And with all the information here, it gives a little bit more description about the gloves he's wearing, the color of the hat, and so on. But I think we can include that for this image. If you actually want to make a similar image, you could have just said make a similar image to my reference image, for example, and it would give you another image. But here I just wanted to test its reading image abilities, which it did successfully. Now as a separate step, I can use GPT or I can go to being AI here. I'll just put the description here and let's create. Okay, here are our results. We do have a line, we have a ringmaster with a hat. The only thing it didn't do is showing the lion playfully bites his head. Let's see the others. Well, this one is a bit better. Yeah, here's how chat TPT can analyze basically any image that you input. It will give you a description if you want, you can use that description in any other platform like mid journey or anything else that you're using. 12. New Update: Examples and Use Cases of DALL-E 3: What can you actually use dally three for? Well, there are so many use cases where dally three will be helpful. And we will explore those cases and the prompts that you can use. Dally Three is such a powerful tool in terms that it's too simple and is accessible to basically anyone. You just get your GPT subscription and you just generate an image of it, will generate basically everything that you need. That's why I think the AI tools are replacing stock images. Because before you just go and search for a stock image, you may find a free one or you would have to pay, I don't know, $510 per image or even more if you couldn't find the free one. Whereas with Dally three or mid journey, you just type a simple prompt of exactly the image you need and that's exactly the time you go and search for that image. In stock images, you get a free with no copyright image that can be useful for your business presentation or anywhere else. I think this is a very important part of Daly three here. Another reason Al three may become your favorite tool is that when you use it with GPT, you actually get the conversation, you give orders, you have conversations and you get the result. Let's move on to other use cases. So you can make logos with it. For a business, a company, you can make book covers, book illustrations, coloring books, card design, album covers. You can make website and product design. You can brainstorm and get inspirations from Daly three. Then you can also make posters, marketing materials, and many more things. The sky is the limit of what you can do with Daly three. Okay, let's try some use cases. So the first one is images for your presentation. Instead of stock images, you can generate an image for your presentation yourself. Okay. For example, here is the prompt, generate an image or realistic potograph. Then you put a specific description of what you want, like a happy person or people in the conference room for example. Then you put that I can use for my business presentation or class or education presentation and so on. For example, this image was created with the prompt, Make an image of a person jumping from happiness that I can use for my business presentation. Because I put the business presentation, it added the details with a formal clothing. Here let's see a different example in Microsoft being I generate a realistic photograph of people enjoying a meal at a cafe that I can use for my presentation. Here are the results that it gave me. I think the one is the best one here. We got two people using an ipad or taking pictures. Then this one is not bad, but it looks a bit fake in the background. So I would probably, if I were making a restaurant presentation, I'd probably use this image. And again, it took only 30 seconds to make this image and you can easily use for your presentation. Okay, the next one is the logo design. Here I would say you can start with being AI because it gives you more ideas. But in order to do editing or if you want to have this development, then you can use GPT here, the prompt that you can use such as design. And then you can list adjectives how you want to see the design of your logo. Like luxurious, simple, vector, colorful. Um, logo for the name of your business, like a cafe, pharmacy and with the name and then just put your name, an espresso club. Here we have the Prompt Design luxurious logo for Spasalon called Harmony. Here we got, we got the name correct. For the other images, the name was messed up with being I you would get four options. That's what I go for. The prompt here, you can see the options I've got. When I run it again and emphasize that I want the name harmony. I got three images out of four with the name harmony where those two spell it completely wrong, then you just need to create more or you can use GPT for GPT because it generates only one image. Well, right now I don't know if in the future you could generate more. But right now they limited to one image per response. In order to brainstorm and get more designs per image, you don't need to wait longer. You can ask it to four designs at once. Here, just make four logos for a construction company hold Skyline. From here you can choose which one you like more and you can ask to expand that design. Sometimes works, sometimes it doesn't. For example here I ask, I like the bottom right logo from the four designs we've created. Expand it and make it in two colors, green and red. Here is the result I've got. I think it captured a little bit from the fourth design, but still is completely different. Unfortunately, when you have multiple designs on the same image, it's a bit more hard to isolate one. What probably you can do is just crop that off, uploaded as the image. Ask GPT to analyze and create a similar design. That's probably the better way of doing that. Okay, a different strategy is that you start with one design. For example, design a minimalistic logo for a construction company called Sky Line. Let's say you don't like a few things. You say, use the image you generated, but use only dark gray and sky blue colors. White background here is the result, changed the background. But for some reason it didn't add the blue colors as I asked. As well as it separated the line, it made a space between sky and line, which I do not want here. Using the same logo, make small adjustments. Make sure Sky line is written together. Use two colors for the logo. Do gray and light blue, white background repeated myself. Let's see the result. This was the result. Again, I didn't take the space out, but it added another color blue. In the next iteration, I ask it keep the logo, but make an adjustment. Make the blue color more light and bright. Remove the space between sky and line. Here is the result. This one is a bit better. We got the blue as we wanted. The space is removed between sky and line. But the thing now, I don't like the second line, I ask. Okay, great. Use the same logo but make small adjustment. Remove one line above Skyline. It did remove it, but now it changed the design quite a bit and added those lines at the top. Let's say you like this design but you don't like the line, then I would recommend to go and use other editing tools that we've discussed, like Clive drop. O to remove the line, that would be much easier and faster than asking it here. Now let's move on to book cover here you can use a prompt like design a book cover for a and then put the genre of the book like Mystery historic children's book or any other novel about describe what the book is about. For example, flying cars or the girl is falling in the rabbit hole and so on. Titled, and just put the title because it could get the name of the title correct as we can see here, The Lost. So here the prompt was, design a book cover for children's book titled The Lost. Let's see some more examples. Here I put the design a book cover for a fantasy novel about a girl who lives in AA society titled The New World. We got different illustrations. Book covers. Those can be good inspirations for an actual book cover. This one is pretty cool. You can see Roberts attacking the world. This looks like an AI body with the Earth. Lots of details. Now next move on to website design. That's another very helpful use of Dally Three because before with Daly two, those things were not possible. Here you can put a prompt, something like design adjectives, colorful, modern minimalist, or landing page for a specific type of website, such as online pharmacy or any other business for website design. The prompt here, minimalistic home page for designer portfolio. I was pleasantly surprised when I saw the results for this prompt. Let me show you all the results. Here are the results first, second, third, and fourth. The reason why I was so pleasantly surprised is that my expectations for a minimalistic homepage matched with the results. This is something if I were looking for a minimalistic website, template would find, here we have pages. But what shocked me is that it actually named the elements it put logo here, web design, UX design. It clearly understood my assignment that I want the website to be a designer portfolio. It knows that he would, let's say my logos here. I would show my web design. Here, I'll show my UX design. This is brilliant. That's how easy you can use Dali three to plan and design your website. Now let's move on to posters. Advertising posters. Or any advertising material. Okay, here we have design a poster featuring. And then you can put person place product such as mountains here with specific text or emotion message. Let's say you want to design a poster that you can tell on C, then you can make it here. This is a poster that I made with GPT and Daly. Here was my prompt. Make a poster with a motivational message. With a few iterations, the text is legible. Here we have believe in yourself even got the correct text. The reason I got it here is because in my prompt, I asked to correct the text, and probably that's why I put it here. Let me show you. Here is the first image I got for the prompt. Make a posture with motivational message. Then what I ask, please improve the text. There are mistakes on the image. Yourself is written as yourself. Here is this result that I showed in the presentation, but it added this correct text, possibly, because here I ask it to improve the text. Then I asked it to remove correct text and remove the background here, is the result a completely different result. I didn't want to go that path. It's easier for me just to remove that line myself. For an advertisement poster, you can just put design and advertisement poster for a specific product such as Heads. Let's see that as well. If I go to being I, this is the prompt I use. Design a futuristic advertising poster for headset. I've got pretty cool images, very futuristic actually, this one. If you are making product photos, then this is something that you can incorporate. Maybe you can even like Photoshop out this part and add your product instead. As you can see with Dally Three, you can pretty much generate any image that you can use for work for. So for example, a school presentation for a family gathering, for example. Dally Three is a simple tool, but yet it is so powerful I encourage you to explore it and start creating. 13. New Update: DALL-E 3 New Parameter Gen_ID: Hello everyone. This is a small update on Ali. I recently discovered that there is a parameter that enables to do editing so much easier. That's the generative ID. You can actually ask when you generate an image to give you the generative ID of the image. The generative ID, let's see what it is and what are some use cases for it. Nid refers to the unique identifier assigned to an image. Each time an image is created, it's given a NID so it can be referenced in future and they identify ensures that if you want to make modifications, references then you can do so accurately. Basically, Jen ID is very similar to the Seed in stable diffusion for example, I ask GPT to generate an image of a couple sitting on a bench in a park and include NID with it. It generated me an image, it gave me a NID. It is what we know from the seed, from stable diffusion, is that if you give the same prompt and the seed number, it should generate the same image. Let's test it out. I'm putting the same prompt here and I'm giving it Jen ID. But as part of a prompt now, let's see. Now as you can see, we're getting the same image and the functionality of Jen ID is pretty much the same as the seed. That would allow us to make small adjustments to the image by referencing the ID. Let's see this example here. I say generate an image of a cartoon character in the children's book style, a girl with curly hair, the explorer, and it gives me an image. Then I ask, what's the end of the image? It gives me the end of the image. Now what I say is keep the image and I put the ID number the same, but make facial features more picture like. Now it knows that I want something in a similar, let's say style. Now it generated this, even though I was looking for more similarities between this and let's say this image, but close enough so we still see Explore Clothing. Okay, let's try one more. Now, this image has a different en ID. So I can ask what's the gen ID of this image? So it gives me a number. Now I say generate exactly the same image, but this time the girl wears a blue scarf. Here is the image. Now what I say is generate in exactly the same image. And then I reference again, not this one but this image. Now instead of blue scarf here I put a red scarf. Look what, Now we pretty much get the same image with the blue scarf, but now with the red scarf. We referenced the same ende, the same prompt here, girl wears a red scarf and girl wears a blue scarf. The end is the same. It has the same starting point and similar generation process. That's why we're getting pretty similar results. Now let me show you another cool thing you can do with end. It's called cross reference or combination. Basically what you do is you give end of one image of the second image and then you ask GPT to combine them in one image. Here, let me show you now I ask to generate an image of an adventure landscape. This is one version then here I like this image, but I think the style is a bit different to my character's style. I ask Jug Pit to generate an image of an adventure landscape with mountains in the style of and I give this look at this style I think matches a little bit better with the style of our character. I ask for end of this image if you want to get Jen ID with every image. Then you can make sure to include NID for all generated images in the future. That's just going to simplify everything. You wouldn't need to ask ID for every image it generates. Here's the prompt. Now generate an image that cross references here. I give the end of the mountain setting of this image here and of our character, this image here here. It generated two images. Usually it generates only one image. But this time it unexpectedly generated two images. My thoughts, the reason is that here I referenced two images and maybe that's why it now generated two results. That's my explanation of what's going on here. As you can see, the images are pretty similar to one another here. Slight modifications but pretty much the same composition. Now, I went on to say, make a full body shot. Here's the result I liked pretty much everything except the facial features here. And I've said improve facial features here. I referenced this image to be more similar. Here, referenced our first second image here. Here is the result. Now I ask GPT to put this image in tropical setting. Here is the result. Now you can see some resemblance of this character with this character, even though there are some small changes. Okay, another thing I wanted to experiment with is to put character in the setting that I give. I have uploaded this image and what I asked for the prompt, create an image that places the explored character here. I reference this image here with a similar attic setting as on the image I've uploaded. Hopefully it would this character to the attic setting like on the image that I've apploaded. Let's see the result. The character features here are quite similar to the reference character, but the details are quite different. I would say that this character is way younger than the reference image than this character. As you can see, it's not perfect. You would still get differences, but it's way better than just using the description and text without the gen ID. With gen ID, you can also experiment with some keywords such as combined blend, merge, because that may give you different results. And booky words like style, aesthetics, design element. Let's see. For example, I've asked GPT to generate an image of a cat in Impressionist style. Here is the image now for some reason it didn't give me a NID. I ask ask to generate an image of a dragon in a modern digital illustration style. Here is the image, I get the ID. Now I want to combine those images. I say use the aesthetics. Here is Gene D of the cat image and Gene D of the dragon image. Let's see the result. As you can see, it used the cat with the dragon as here, the colors I think are from the dragon. The style is the impressionist style of the cat. Okay, it's not really a fuse, but we see cat and dragon side by side. Again, here we have two images. Okay, now instead of use the aesthetics, I say combine elements. Here's my dragon image and cat image. Let's see the result. As you can see, the results are a bit different. Now, the dragon and the cat are side by side. Here the dragon is in its pallet, in its own original calpalate. Let's see the other one. And here we have also predictable style. Now it's time to experiment with different prompts and use cases of Jen ID. 14. Prompt Writing - Subject and Medium: Hello everyone. In this module you will learn how to write prompts. This will be applicable to all of the AI tools that we cover in the course, but especially to the stable diffusion based tools including images, Do Lexica dot Art, Dream Studio, blue willow, Leonardo Astra, and Automatic 11 11. I'm going to do the prompt writing in a tool called Automatic 11 11 which is a stable diffusion program. We will be covering this program at the very end of our course because it's a bit more advanced. But for prompt writing, I decided to use this one because it gives me greater flexibility. But again, these lessons could be applied anywhere. At this point, you don't need to follow along with what I'm doing because we will use all these concepts in the AA tools we cover next. However, if you'd like to, you can jump to automatic 11 11 section on how to set up and run Automatic 11, 11 so that you can follow along with what I'm doing. But again, that's not necessary at all. Before we begin, I just want to outline a very great resource for prompt writing. It's a stable diffusion prompt book and it's brought by open art. It's a guide into prompt writing and it goes into a lot of details as well. A gives a lot of interesting examples, artists names, and so on. Throughout my presentation, I will be referring to some of the content from here. Let's begin by writing a short prop, for example. We, we'll usually start with the subject. For example, a man. Let's click Generate. As you can see here, we have four different images of men. In this case they were pretty similar in style. However, if I do it a few more times, you'll see that it can be completely different. Here we do write any specific details, it's up to AI's interpretation. It has a lot of room for imagination. Here on some of the images, I notice that the head is cropped. To fix that, we can use the negative prompt. Negative prompt helps AI to know what we do not want to see in the image. In this case, I can put crop and crop head. These are some things that I do not want to see in the image. I don't want to see cropped image or crop head. Let's try with this negative prompt. As you can see here, we got another four images of men. The three of them I think are pretty fine. Those three, but this one is cropped just a heads up. Even though you put it in the negative prompt. Sometimes I would not quite do exactly what we write at the prompt. The only way to fix it is just generate lots of, lots of images. And another way is to actually write a more longer prompt, which we will do. We started with the subject. Subject can include like people, person, man, woman. It can include animal. It can be some object glass of water. For example, Castle. If you want some landscape sunset. Also we have here celebrity names. For example, I can put M. Watson and generate a few images here in the negative prompt. I would also like to put not say for work and naked, so we don't get any naked images. Here we have it. We have four images of Emma Watson. You might be wondering how does AI know about Emma Watson? Well, A I was trained on a large data set of images that were publicly available. As you know, there are tons of images of celebrities available online. Ai knows most of the celebrities pretty well and you can use it in your prompt. We can also try glass of water just to see the objects. As you can see here, we've got exactly what we ask for, a glass of water. As you can see, different styles in order to build on that. We can now specify medium and art style for medium. We have oil painting, watercolor, photograph, pencil drawing, airbrush, digital art, technical diagram, three D, illustration, vector, and much more. For art styles, it refers to a historical art style. We have abstract Renaissance. This would be a. A style of Leonardo da Vinci. Mona Lisa, for example. Impressionism is Bango cubism, contemporary pop art. And then we have surrealism and fantasy to add to glass of water, we can put water color color painting of glass of water. That would narrow down what style AI uses. Here we go. We have four images of glass of water in watercolor style. As you can see, it matches pretty well. And also the color scheme, for some reason is very similar to, we can also try photograph. I know from experience that if you put photograph of a subject, for example, of a woman, it would tend to be black and white just because the photograph is more predominantly with the older photographs. You can see here that all the images are black and white. And that's just because AI has association of the photograph with black and white photos. Let's say you want to make a modern photo of a woman. Then you would need to put some adjectives, for example, modern photo of a woman. Let's see if that does it, It may not work. You can see here that the photos we've got are still on the black and white scale. They don't seem to be modern at all. To solve that, I'll just add a little bit more words to the prompt and to the negative prompt. To start with the negative prompt, I will add black and white. I don't want to photos, I don't want monochromatic. Monochromatic one color which is black and white. For mode photo a woman, I would add woman in a T shirt. There are items that belong to a particular period in time. For example, T shirts, jeans, heels will be something very contemporary. When you list specific things that belong to a specific time period, you'll get images from that time period. Let's generate. You can see here that all the photos here contemporary, you can see a woman in a T shirt. That's exactly what I've said. If you want certain time period, make sure to include, um, items of clothing or accessories from that time frame. Okay, that's for medium. We can also try art style. I think style. Let's change the pro, let's try abstract drawing of a car. Here we have again four different images, the first two and the fourth one. We see clear drawings here in this prot, I actually combined the medium and the art style abstract drawing. And you can try out those things, You can combine, mix and match the art style and medium together. Let's do one more medium, Let's do a technical diagram. You can see how changing the medium affects the image. Technical diagram. Okay, let's see technical diagram. On the first and the second one, we see labeling different viewpoints of the car. That's what we would expect on a technical diagram. Let's do, let's do pop art art. Here you can see that the images are in completely distinct style compared to the previous batch of images. This one is the pop art and again, very big effect of the art style on the image. 15. Prompt Writing - Composition, Action and Details: In the last video, we've started writing prompts. And we've started with the subject. And then we went further to talk about medium and art style. And how you can use specific words like well, painting to define in what style you want your images in. And that has a very big effect on your output. Now I would like to even further add to our prompt and I would like to talk about composition. So there are two types, There's short type and point of view. For short type, it's basically where do you want to see your subject? Is it like a close up or further away? You want to see a full body of your subject for point of view, imagine a photographer takes photos of a castle, and this is basically where the photographer would take his photos. Is it from a low angle shot or it's from a drought? As you can see, these all have a big impact on the image itself. Let's try some of them. Let's do a close up portrait first. On these images, this is what we see. We exactly see just the face of a girl and maybe a little bit of shoulders. What would the close up shot be? Now let's also do a full body because this one is a bit of a tricky, the images we've got are quite, for the best one is the first one and the second, but the second one is not quite full body but the rest are cropped. So you can see this one, the head and the shoes are a little bit cropped here as well. In order to not have those cropped images, I can add specific words, definitely to see the hair. For example, I can add hair or a hat to make sure that I doesn't cut through the body. I'll add the hat here. Also, because I want a full body, I can add specific elements of the body that I want to see. For example, if I want to see legs, I can put legs or certain attributes of clothing. For example, for full body, I can also put shoes or boots. I'll put those things. I'll put a hat, and I will also put boots here. This one looks a little bit better if you want to see a person or your subject a little bit further away. It would be also nice to add a background and we will talk a little bit that, um, in our next slides I'll just put in a park and we'll see how that will change the images. I will also put a photo, full body photo of a woman had boots in a park. I think this is a little bit better in terms of composition. As you can see, the subject is placed a little bit further away and we can see boots and the head. And basically everything except for this image where AI just ignored the boots and didn't render anything here in terms of faces for AI. The further away the subject, the less space it gives to AI to properly render faces. For now, disregard the faces. Okay, here's the full body photo of a woman. Now let's try some point of view. Let's do, for example, we angle shot of a castle. Here's some images that you can expect. When you use wide angle, you'll see your subject in full. Usually, I found that wide angle view is somewhat similar to panoramic view. Now let's change this to low angle shot and see how that will make a difference. Low angle shot, let's do a photo here. Compared to the previous one. Most of the images we see the castle above. And that's what low angle shot means, that the camera, if we were talking about the photos, is located below the subject. Now let's compare that to a high angle. Here we can see the subject below. All of the images are consistent and some of them I would say are drone photographs of the castle. You can also use drone photo of castle. To finish this up, I would like to use fish eye photo of castle. Here we can see different images, but as you can see most of them are distorted. This is what we would expect from a fish eye. This one I would say fish of a ceiling in the castle. Maybe not sure. But these three I really like. Now you can use composition words to define how far or how close you want to see your subject and from what angle. Now let's go further and talk a bit about action. Action can be very important for certain images. For example, for dynamic images. Action images like soldiers attacking a castle, for example. But even simple images for a portrait, action can be important. For example, when a photographer takes photos of a model, he usually directs where she should look or how she should stand. Similarly here, you can direct to how you want your subject to be. For example, you can say, a portrait of a woman looking up. Let's try that. As you can see here, compared to the portraits we tried before, where the face is usually looking straight at you. Here, we can certainly see that the woman is looking up, her head is tilted, And this one as well, I would say the second and the third ones are my favorite. Here, you can see how you can use the action to impact the posture of your subject. This one was for looking up. We can do, for example, reading. Let's do a cartoon of a boy reading a book. Here you can see a boy doing certain action, which is reading a book. I certainly captured exactly what we asked. So that's it for the action. Now let's add more details to our subject that can be very important if you want your image very detailed. For example, if you make images of a woman, you can ask yourself questions. What does she, does she wear jeans? Or does she wear a dress? Or is it a historical persona, for example? Then you can specify certain attributes of clothing of that time. For example, a tunic or a corset, for example. You can even add more details to that clothing item. For example, what kind of dress is it? Is it a short sleeve, long sleeve, puff sleeve. Maybe it has embroidery or lace and stuff like that will add a lot of details to your image. Also the hair you can specify what hair does the person have, is a long wavy hair or maybe a short and dark hair. For jewelry, you can put, for example, pearl earrings, bracelets, necklace and so on. For shoes, you can put sneakers by flats, hiking boots, heels, and others and accessories. You can put things like a scarf, sunglasses, maybe a heat or handbag and so on. So let's try some of them. Let's do an oil painting of a woman and then I'll put dress. I'll put floral embroidery and lace. And then puff sleeves maybe also I'll put that she has pearl earrings. Let's try that. In these images we see a lot of frame. Actually, I don't like that. I'll put that in the negative prompt. I don't see any frames frame. However, as we specify that it's an oil painting, usually oil paintings come with frames and exactly what AI generated. If we zoom in a little bit, we have some maybe pearl earrings and you can see the puff sleeve stress place. Every detail that we've described in the prompt is presented here. As you can see here, you can use your prompt to specify the specific details you want to see in your image. 16. Prompt Writing - Negative Prompt, Stylizers & Modifiers: In the last two videos, we've talked about prompts and how you can use specific elements to get the desired image. We started with the subject and then we talked about medium and art style. We've talked about composition and action, as well as the details that you can add to your subject. Now I'd like to talk about negative prompt. Here I've compiled a list of words that I think would be very suitable for negative prompt. These are the things that I do not want to see in my image. For example, it is bad framing. Out of frame, bad anatomy, bad proportions, blurry crop staple diffusion makes mistakes with how many arms people have. For that not to happen, I also include extra arms, extra fingers, extra lex, and so on to make the face and hands more detailed. We can also put poorly drawn face or poorly drawn hands. Then we don't want any text signature waterworks and so on. I'll just copy this negative prompt to our program here. This will actually help to make our images look better because some of the arrows or mistakes can be avoided with negative prompts. Okay, now let's talk about background and environment. This is where you want to place your subject. For example, it can be a man in a park, or a dog in space, or something. Underwater. Underwater is a nice one. Let's try grocery store. Let's do a portrait of a woman in a grocery store. Okay, here we've got images of a woman on the background. You see the grocery store here, we have some fruits here that's captured our background. Sometimes it helps to add background, it knows exactly that this refers to the background and this is our subject. Let's move on to the next element, stylers and modifiers, stylizsre, words that modify the look and feel of the artwork. For example, lighting. There can be a big difference between a moon light versus a daylight or a studio light. Here I've included soft, diffused light, sharp street light, moonlight, cinematic studio lighting, morning sunlight, natural lighting. Here I'll put a portrait of a young woman, contemporary dress, neoclassical style, place embroidery in a magical part background. Now I will add the lighting, moon light. Let's check out these images. I think the first one is set in the dusk. This one has a little bit of sunlight, but the other two, this one and the next one, it does look like a moonlight, maybe also a little bit of a street light. Let's compare that to daylight. As you can see here, these images feel very different to our previous batch of images in the moonlight. Here we can see a nice day light, that's the effect of lighting. Next is the color scheme. This is what colors you want to see predominantly on your image. You can put motor black and white or vapor wave is a certain style vapor wave. You can have this style of colors. Also, you can specify maybe it's called warm colors, vivid colors. Pastel colors. For example, let's try pastel colors, daylight pastel colors. I think that will work very well with the style. Let's see these images here. We've got color scheme. As you can see, it's quite pale and soft, there's not much contrast. That's what the past color scheme is about. Okay, let's move on to resolution. These are actually super important. If you want a good quality image, you should always put any of that highly detailed, intricate HD R 64. It's basically just the resolution, we can put that in our image. 64 detailed, as you can see. By adding highly detailed and resolutions such as 64 K, it added a little bit more details in the images to look at the slays. Next we can put specific words to make our image more realistic. For example, keywords like Unreal Engine and octane render. You will usually see in other prompts to make images more realistic. Unreal Engine is a real time three Z creation platform, usually for games that makes images more detailed and realistic. Octane render is a rendering engine, specialized photo realistic rendering, realistic three D scenes and lighting stable diffusion knows about these keywords and will actually produce better images. You can also use hyperrealistic, ultra realistic, and photorealistic. Let's try to use Unreal Engine with our prompt Unreal engine. As you can see here, the girl stands out a little bit more here. And just the way the lighting works makes her figure looks more real. These words, the stylizers, highly detailed, unreal engine, and the lighting will all make small adjustments that overall make the image way better. Now we can go to emotions and adjectives. This is how you want your image to feel. So it can be like magical, romantic, or it can be gloomy horror epic. In my prompt here, I already have the magical background. Usually for adjectives they don't apply only to the specific or park background, they actually apply to the prompt and to the full image. If you put like gloomy background, then the overall feel of the image will be gloomy here. I wouldn't add any extra words, although I could put like fantasy and so on. Okay, that's for emotions. Next, if you're doing a photo or photo realistic image, you can add specific keywords that would apply to photography, such as aperture. You can put like 1.8 which is great for portraits. You can put ****, for example, 80 millimeters or macro. If you want an image of, I don't know, an insect for example. You can put specific camera, for example, phone camera would make a different image to a professional cannon photograph. You can also put long exposure if you want, um, night lighting effect. With that, let's put aperture in our prompt 0.8 and maybe 80 millimeters with the aperture of 1.8 and 80 millimeter ****. I would expect the background to be blurry and the person's figure, but very detailed. Let's see if this is what we've got here. Yes, I can see on the first two images that the background is blurry and we've got this intricate details of the dresses here. Next we can indicate websites. In our propped for example, we can say trending on Art station. Art Station is a platform for modern illustration. At the time of training, standard diffusion would know what was trending on that. Platform sieve is a Japanese anime style, this is the Platform Pit Net. If you're creating anime style images, then you can put trending on Psi. Then you can also include like Instagram and so on That will give the photo more modern feel. We can trending on Instagram here you can see that the posture just feels a little bit more modern, something that you would maybe see on Instagram. I highly recommend using stylizers and modifiers in your prompt, especially the ones like resolution and realistic keywords, because they actually make your image look way better. 17. Prompt Writing - Artists: So far for prompt writing, we've covered the subject medium and art style composition, action subject details, negative prompt background, and we've finished with stylizers and modifiers. The final element for prompt writing that I'd like to talk about is artists. Artists may have a very strong influence over the image because the influence in what style you will get your image. For example, here I've listed three artists, Alphonse Mucha. If you add that to your prompt, you will likely get a two D illustration for Frida Lo, you would get a mix of surrealism, symbolism, and modern art. For Ing, it will be an impressionist style we can try with our prompt here, I will first remove something that will conflict with the test. For example, trending on Instagram. I'll remove that. I will remove the photograph, the aperture, because that's for a photograph and I'll remove a real engine. I will keep highly detailed and I will remove as colors because I want the color scheme to be influenced by the artist here. I'll put, okay, let's see the images. As I can see here, all the images repeat the style of Van Go. Even the brush strokes, this is something that Van Go used in his paintings. Except the hands here, I think. I didn't know what to do with the hands. It just used more or less photorealistic hands. The hands look out of place. Similarly here, the face and the hands are a little bit out of place. Okay, now you know that artists have a strong effect on the image. I would also like to say that you can actually add multiple artists here. You can put let's say alpha. I'm not sure what it's going to be, probably merge of styles here. We can definitely see that the background and the clothing item on all these images actually is Bangor style. However, look at her face, neck and hands. That's in my opinion, that really looks like Alphonso style. If we go and Google Alphonso, I found some images here and look at the face. In one of his illustrations, I think A, I tried to capture some of this facial expressions and details of Alphonso style. Here's how you can combine the different artists. You can even have a third or fourth artist here if you want. Another great reason for using artists in your prompt is that certain artists will help AI generate correct proportions and correct faces and hands. Ai still struggles with faces and hands. Having an illustrator like Alphonso or digital artists where in their works they have very detailed pass, having them in your prompt will actually really guide AI into producing better faces and better facial features. Okay, where can you find those artists? If we go to open art book, the one that I showed you in the first video here, they list a few artists here for different, like portrait artists, landscape artists, horror artists, anime scifi. However, this is a very small list. I found a few resources for you where you can go and look for the artists. The first one is a prompt guide. If we go here, you'll have the name of the artist. Here are some images that were generated with stable diffusion. As you can see, different artists will have different images and styles. You can go and check it out there. Also, you can search for the specific artists here. Or if you are doing it in a specific style, you can go and choose the category, for example, painters. And you'll have all the artists that are in this category. Okay? The next one is this website. I actually like this website a little bit more because here you have more artists and it's easier to see what each artist is best for. For example, if I want more detailed pass, I would probably use this artist. You can click on it. You'll have more examples here as well. You can just copy the prompt. You just click on it and the name of the artist will be copied here. You can see there variety of styles and artists. This one is also a very beautiful one. I'll copy this one and try it with my prompt. A highly detailed I will remove the Ang, I'll put Alphonso and Emily bald. I think the Alfa. And this artists will go well together because they are both illustrators. Let's see. Look at these amazing images, look how detailed all of them are. The face looks stunning, I think because we've used two illustrators. I made faces look way better than when we tried it before. We have dress and look at the park background that AI tried to implement with these artists. I think we achieved really good quality images. Okay, a third website that you can also use is the screens notion here as well. There are a bunch of artists that you can use and check them out. Okay, now you know where to go to check artists or look up for artists. 18. Prompt Sample - Portrait: Now I would like to summarize everything that we learned so far about prompt and talk more about the order. As well as actually do the prompt from scratch to finish. And show you how I work with prompt to achieve the desired outcome. Okay, the order is actually very important because AI pays more attention to the beginning of the prompt and the ending of prompt. If your prompt is very long, it may Mrs. words or concept in the middle. It will put more emphasis in the beginning and the end. If you have certain details that you want in the image, make sure you put it at the beginning or at the end. I will also show another way how to emphasize certain words to have them in the image. But for now, let's talk about what usually should go at the beginning and what usually goes at the end. At the beginning, we have the medium because that has a big influence on the artwork. Then we have our subject, action and details. This is all about the subject. What is he or she, what are they doing and details about the subject. Then we have the background and stylizers, words that describe lighting, words that improve resolution of the image, and so on. At the end, we have artists, I've made the medium and artists in the same color, because in a way, artists do influence the medium and style of the image that you will get. Now I want to show you my process for prompt writing. I'll start from scratch and we'll improve prompt until I have the desired image. First, I want to start with young woman. I want to create an image of a woman in ancient Egypt that looks like Cleopatra. We'll try to do that here. Young woman, that's my subject. Now for medium, I want air brush painting. Airbrush portrait, a portrait of young woman. Now I want to specify the details of the clothing. She will be in tunic and she'll have gold jewelry. Let's try to generate this here. We've got some images right now. They're far away from my desired outcome. I will keep going. I want to specify gold jewelry I want, and I want gold earrings. Now I'll put the background. In ancient Egypt, temple with columns. With columns background. Let's try this. These images are a little bit closer to what I'm thinking about. However, I don't like the style here. We'll keep on working what I like that now I can clearly see more Egyptian style here. That's because we have ancient Egypt here. And that's an important keyword. Yeah, now these women look like Egyptian Pharaohs. Okay, now let's add stylazersIillutlightI. Want a dramatic light, dramatic light, high contrast. I would also like to add Unreal Engine and 64 K, highly detailed. Let's see, now we're getting images that look like a three D character. It doesn't look like a photo or painting. That's, I guess the effect of Unreal Engine because that what is used for games. And here we can definitely see like a three D character that can be used for games. What I will take Unreal Engine out of here, I'll just keep the 64 K and highly detailed. I will also put here hyper realist. Hyperrealistic. Also, I want the woman in Egyptian ethnicity. I'll put airbrush portrait over young Egyptian Egyptian woman. I could also put a historical figure like Cleopatra. I could put Cleopatra. Maybe I will specify that I want tank tone tank. Okay, let's try this. Okay, we've got some interesting images here. I like the scripture background. I actually add that as well to my prompt. See, the hand is bad here. Here we have a frame. This one, I like this image, but it's off center. Let's try again, and let's add a little bit more details. Now, I'd like to add artists. I've decided to use Greg Rutkowski which is a modern illustrator. I can go to our stable diffusion che sheet and paste the name and search. Here you can see some styles that were created with Greg Rutkowski style and you can see that it's very detailed and that's what I want in my image, I will use him. By Greg Rutkowski I also want to add Alphonso much because I really like his style as well. Let's try. As I said, I liked the scriptures on the columns. I will put that somewhere here. Columns with scriptures. The images I've got here, I've got a full body shot that was cut, cropped, I've got a picture here. None of the images here I like, I think the lighting, the gold looks a little bit fake. Instead of just gold, I'll put iridescent gold, Iridescent gold jewelry. And hopefully that will make the gold color more deep. Also, as you can see, we've got a full body shot. I will do airbrush close up portrait of a young Egyptian woman. Let's try this. Here you can see that the background is plain. That's probably because our prompt God a bit longer. This ancient Egypt template with columns and scriptures background gets lost a little bit in order to emphasize that we can use parentheses. Each keyword has a weight of one. When you put parentheses, for example, I will put temp, I'll put parentheses here. And this will make the keyword weight of 1.1 It adds extra weight and gives AI a flag that make sure to include these words. I also will highlight columns with scriptures background as well. Also, I would like to add aperture focal point of 1.8 to make sure that the background will be a little bit more blurry. Also, I would like to add some emotions. I want it to be epic. I want to highlight it, There's two ways to emphasize it. So you can put parentheses. Another way which is equivalent is you can put it in parentheses and then Cullen and put 1.1 That's basically the same as just putting it in parentheses. However, now we have an option not just to do 1.1 we put a different weight. We can put 1.2 or 1.3 which we cannot do just with parentheses. Let's put epic 1.3 okay? Let's try this. I think these images came out really well. We will need to fix the eyes a little bit, but that should be fine. Yeah, overall, it looks nice and I like the neck here, very intricate details. Overall, I'm very satisfied with this one. I will run it and improve the eyes a bit after face here. The images that we get, I think this is impressive. I would probably this image, maybe I'll correct the small artifact on the head piece. But overall, it looks astonishingly good. Okay, this one is not my favorite. This looks more Indian style. Interesting. Okay. I would keep this one. I would, I would save it, but just for fun, let's try a few others. And as you can see here, we have the columns, but we don't have the scriptures. I will remove parentheses here. I'll put parentheses on the scriptures. I will put a higher weight. I'll put 1.2 tunic. I've noticed that most of them are not wearing a tunic. I will also put a little bit more weight on the tunic. Let's put 1.3 for example. Let's see these images, these are fantastic as well. I like the third one. Let's try to generate a few more. Because none of these have columns, I will put even more emphasis. Ancient Egypt temple with columns and I'll put parentheses here. Another trick is that instead of one parentheses, you can use two parentheses here that would be equivalent to, to this one parentheses, one parents. Then we put column and then 1.21 Basically it's 1.1 times 1.1 which is 1.21 Or you can use two parentheses, that would be the same weight. We will cover weights in a little bit here to make more emphasis. You can use two parentheses or even three parentheses. Yeah, that will give it more weight. Again, some stunning results here. My favorite one is the first one. We have all these details of the scriptures in the background. She has beautiful face and earrings. Well, this one looks a little bit longer. I'm not sure if there's any symbol to that, but that for sure can be corrected. I think with this prompt, we achieved a great results. So we can stop here. There are other ways you can improve the image. But this is a little bit more advanced. And we'll cover that a little bit at the end of the course when we will talk about this program. But for now, we achieved everything here. 19. Prompt Sample - Landscape: So I quickly want to show you how to create prompt for a landscape. This was for a portrait. Let's quickly do landscape. Okay, again, I start with the subject. A castle. I want a wide angle shot. Wide angle shot of castle. Now this castle, I wanted to be a mean evil castle. I'll put evil. Now I want the castle to be in a forest. I can, in a forest I can specify what is that I want in at want a color. Flowers and trees also for the medium. I wanted to be an oil painting. Oil painting. A white angle. So maybe I'll yeah, Oil painting and then I'll put white angle shot of Castle Devil in a forest. Let's put Medieval Castle. Do a little bit of rearrangement. Medieval castle in a forest. Colorful flowers and trees. Okay, let's see what it'll generate. Here we have our images. Yeah, it's beautiful. Now, I want to make this castle feel a little bit more magical. I will put adjectives magical. I will put epic also to make it more like a fantasy castle. I will make it flow in the sky. I will put the action here, Medieval castle. And I'll put action floating in the sky above clouds. Now it's not in a forest, then I'll just move in a forest. Forest will be just an element for the image, not the background for colorful flowers, trees, magical epic. Let's also add stylish, realistic, and detailed. Let's create this now. We can see some here. Maybe I tried to put some clouds. Although here it looks like F doesn't look like the castle is floating. Yeah, none of these images look like the castle is floating. What I'll do here, I will emphasize this action. It means evil castle floating in the sky of clouds. I will emphasize this by putting a double parenthesis. Also, I want to add artists. I already chose two artists. It's Adrian Everson. Let me show you who he is. Our Chet, here are some examples of his artwork. Again, very detailed. I also have a gurney. Let's add him here as well. Here is not found here. Let's try, maybe here. Okay. Now here you can see that colors are nice and soft and this is what I want to see in my images. I will use his name as well. Let's go back to our program. I will put those two artists. James Gurney. I'm Adrian Everson. Okay. Let's generate and hopefully this will work. Okay. Let's see. Wow, this looks way more magical. It still looks like a fog but now, because we added the artists, I love all those details here. Yeah, some fog and clouds and looks like the castle is in the sky. To make it even more clear to AI that I want castle floating. You know what, I will remove the white angle shot. That might confuse AI a little bit. Beneath is forest and with colorful flowers and trees. I'll put three parentheses here. Three parentheses in the sky above clouds for details. I can also put more resolution. I can put 64 K. Remember we also mentioned that we can put things that are trending on certain websites. You can put trending on Art Station. I think that may make our cast a little bit more fantasy like because Art Station is for modern illustrations. Also I can put fantasy here. Magical and fantasy epic. Let's emphasize fantasy. Okay, let's strike this out. Okay, let's see these images as you can see here in the first one first image, the castle and a little bit of forest are floating on the Broke there above the clouds. I think AI captured our prompt very well. Here again, you can see some clouds in the forest. I think this is beautiful overall. My favorite one is the first one, and this the first one something that I would keep. If for some reason you're not getting your desired output with a long prompt, try modifying it a little bit. If that still doesn't work, try just generating a lot of images because then there's high chance that one of them will be something that you're looking for on this node. I think I've covered everything that I wanted to tell you about. The only thing left for me to explain is the keyword weight about all the parentheses. Again, there are two ways you can create higher weight for your keyword. You can put the keyword in parentheses and use keyword column, then just put the number. It has to be one point, something. This increases the keyword strength. It has to be higher than 11.2 or 1.3 Sometimes you may want to decrease the strength of a keyword. For example, colorful flowers and trees. I want less of this in the image. I can put parentheses here. We'll put a column, and I'll put a number that's less than one and higher than 00 point. Let's say eight, that should give me less of the colorful flowers. To decrease the keyword strength, you would put the keyword in parentheses and you would use column and put a number that's less than one and bigger than zero. For example, 0.9 That decreases the keyword strength. It has to be less than one. Okay? Another way you can increase the keyword strength is just using parentheses. Just putting the keyword in parentheses, that will increase the keyword strength. If you put only one parentheses, it will be 1.1 If you use two, it'll be 1.21 If you use three, it's equivalent as like putting keyword column 1.33 To decrease keyword strength, we can use brackets, one bracket. The keyword in a bracket would have the weight of 0.9 The keyword with two brackets will have a weight of 0.81 and the keyword with three brackets will have the weight of 0.73 That's for keyword weight. 20. Prompt Writing Resources: Now I want to talk about resources for prompt writing. First of all, it's an open art prompt book that you've already seen. Here are a lot of examples. They explain different keywords. For example, for cameras like drone thermal camera footage, there is a great stable diffusion art guide that also has keywords here. For example, style hyperrealistic, the word hyper realistic, and here's the node. It increases details and resolution. I think this will be very useful for you if you are just starting out and unfamiliar with some of the terms here. If you want a little bit more information about prompt writing, you can use the skid here. They actually write a prompt from the scratch and would explain every step of the process as well. Next is a list of modifiers. Here, if you go to this website, it has keywords and you can see what images can be created with these keywords. For example, dim light or light diffraction, studio light, those different things here at the top, actually, right now it's lighting keywords, but now you can change, for example, to effects. Here you can check out some keywords like bulk effect or neon light effects and see what can be done with that. Filters, lenses and so on. This is a great tool to check out. Next is mid journey styles and keywords. Even though it's for me, journey, it has a keywords, for example, artists or materials. If you click on materials, there are a lot of keywords. For example, solids. It will help you to find the vocabulary or keywords for your prompt. For example, wooden or lumber, sawdust, and so on. These images are journey generated images with stability diffusion, you may not get the same images, but it's great to help you find the right keywords for your prompt. Another great resource is Prompt Hero. Here you can see a lot of beautiful images generated by others. If you go on the top here, you can see the featured images at New Top. Then you get specific platforms. For example, mid journey, you'll get the mid journey images or Dali. These are all images that were created with Dali engine or stable diffusion will be all the images created by stable diffusion. If you want to save any images that you liked, you will need to create an account with prompt hero. Then you can browse through different images and actually save them by hitting the like. That will be in your profile. In your favorites. Here are the ones that I chose and they are in my profile. I can always go to click on this image, for example, and see what prompt did the other person. So this is the prompt, this is the negative prompt. Here's the generation parameters. We will discuss this later and model used stable diffusion 1.5 for example. For stable diffusion, I like this one. See how long this prompt is. But with the information that I gave you now, you can identify why did the person chose those words and how did he or she structured the prompt? For example, where's our subject? It's gorgeous Norwegian girl. And what is the medium? Professional portrait, photograph. And then we get the details in winter clothing with long wavy blonde hair. Look, freckles, beautiful, symmetrical face, huge natural make up. These are all details of the subject. Now we get to the background, standing outside in snowy city street. And you can see here that there are parentheses, two parentheses. Now you know that this is emphasized. And here are our stylizers. Ultra realistic concept art, elegant, highly detailed, integrate sharp focus. Here's our aperture ****, medium shot, volumetric fok trending on Instagram, our websites trending on tumbler, HDR and resolution. And we've got some negative prompt here as well. So now we just can copy this prompt and try it out in our own program. So let's go and just paste the prompt here, Let's check out the images. I think the second one came out exceptionally well, and the third one as well. Even without writing prompt yourself, you can go to Prompt Hero, choose your favorite images, Copy the prompt here, then you can change the details. For example, you like you like this image, but you don't want it to be in a winter time. We can actually change winter clothing. Let's put summer clothing standing outside in snowy CT Street. We will change that just in city street. And this should change our image a lot here for gorgeous images that were completely generated by AI. And we can see beautiful blurry CT Street background, we can save the image. Now you know that you can use prompt hero to get inspired and improve your prompt writing skills by checking out prompts created by other people. Okay, there's another good resource which is Lexica Art. We will actually cover Lexica in one of our modules. I'm not going to go into too much detail here, but again, there are lots of images and you can search for prompts. If I click on this button here, you can choose the model. The Lexica aperture is native to Lexica. If you're using stable diffusion, then select stable diffusion again. If we click on the image, for example this one, I can see what prompt was used to generate this image again, I can use that in my program here and try out this prompt again. You can check out these images and learn from them. Now, I think we've covered quite a lot on prompt writing. And you should be all set to create your own prompts. And start experimenting with prompt writing, happy writing, And see you in the next module. 21. Lexica Introduction: Hello everyone. In this module, we will cover Lexica. Lexica is another image generator, but it's also an image search engine. If we go to Lexica here, you can search for images and you'll get tons of AA images that were generated by others. If we click on one of the images here, you can check out what prompt was used to generate the image, which gets quite handy. Lexica uses its own model called Lexica Aperture. Currently, there are two versions available, version two and more advanced version three. These are both fine tuned models based on stable diffusion. Lexica was founded in 2022 by Sheriff Shamim. Okay. Now let's talk about pros and cons For pros, Lexica has simple interface and it's easy to use. If we go back here, I think that Lexica has one of the best user interfaces. The first window is home, this is our search engine. And the Generate is where we write prompts and can generate our own images. Very simple. Another great feature is that it is an image search engine. So we can go and look for images, get inspired, check out prompts and so on. It produces high quality images. If we just go in the gallery and look through the images, for example, a teapot here, it looks spotless. I don't see any artifacts for portraits as well. Look at some portraits here. And also it's photo realistic images. If we compare it to some other platforms like Ali, look at these phases. If I even zoom in the lighting, the eyes, the rendering looks spotless. It's a great model for photo realistic images. It works well with basic prompt. If we compare Lexica to just like the basic stable diffusion model, to get high quality phase or high quality portrait, we would need to add lots of stylize. Also add artists that help to make the face look better. But for Lexica, if we just write like a woman or it would create beautiful portrait right away here, for example, we see a short prompt and we already have beautiful, beautiful images here. Okay. It also has three limited credits. If we go to the account here, you can see that you've got 100 images per month. You can see how many you've already used. For example, I already used seven of 100 images. Lexica also has image to image generation. Basically, you can upload your own image and Lexica can generate AA images that look similar to the image that you've uploaded. So that's a great feature as well. Another great thing is that it allows private image generation with a paid plan. If we go here, here we can I keep my images private. Images created under the start and pro plans will show up in our search engine. If you subscribe to the max plan, then all your images will be private unless you decide to share them. To make your images private, you will need to get the plan this one, Okay? What are some disadvantages for Lexica? There are only two models available. As I said, a aperture version two, and version three. I don't know if you've already noticed, but the images that Lexica creates it, distinct recipe for all because it's a fine tuned model of stable diffusion, it uses a distinct recipe. It creates images, I would say in somewhat similar style. Just look at some lighting and colors. For example, if I want, let's look at Go. Here we have the style of being. But look at the colors. These are Lexica colors. These are not being colors. The lighting and the soft, smooth ear brush like texture. That's what lexica adds to the images. If you want to create images with white artistic variation, that may be not the best platform for you. It also has limited advanced settings compared to, for example, the platform with covered images I, where you can change like seeds and you can use other models and so on. Here, if you go to January, Advanced Settings here you can choose the dimensions of the canvas, choose the model type, Lexica, aperture, version three of version two, and you can use the guidance scale, and that's it. Limited advanced settings here. Also it requires a paid plan to use images for commercial purpose. I put that as a disadvantage because usually for, for a lot of platforms, you can use images for commercial purpose right away even without a paid plan. If we go to account here, can images for commercial purposes and they reply, you can use any image you find on Lexica for personal use. For commercial use of images created with Lexica, you must have a paid plan with some restrictions on team size. If you're a team of two to five people, then you need the pro plan. Teams of five plus need the max plan. Please see our license page for more details on allowed usage. The information I have on my slides comes from Lexica Art website. However, as a disclaimer, none of the information I tell you is a legal advice. So make sure you do your own research or consult a lawyer for legal issues. That's it for the Lexica introduction. In the next video, we'll go over some functions and we'll try different prompts and generate different images. 22. Lexica Features: Now let's check out what Lexica can offer us. First, I want to start with the search engine. If you go down here, you can see a whole gallery of AI generated images if you like any of them, you can go and put this heart, let's say I like this Capybara. I'll put a light here. Every time you put a light, it will be added to your light gallery. Here are all of my images that I've liked. You save styles, if you find something inspirational, you can save this for later. If we go back to home here, let's say you're interested in a specific style, you can look this style up. For example, pop art. Here, you'll have all the images that have this keyword. You can also look for specific objects, or for example, if you're interested in some products or looking for inspirations. For example, cream here, let's put cream product here. You'll have all different cream products. For example, this one minimalistic photo of natural cream for skin care. This one looks lovely. Again, if you liked any of the designs, you can save it for later. Or if you like this style, you can click on this button. Explore this style. That will just search for images in the similar style. For example, this one is very interesting. Let's try a different one. Not sure what this is. A herbal supplement, capsules back surrounded with nature. Okay, For example, this one, let's say you like the overall look of this. You can go and open this in Editor. Now you can add some, I don't know keywords. For example here it says iphone photo of natural cosmetics, Flower serum, Camp handmade cosmetics. I will change this a little bit. I'll put just Photo natural cosmetics. Just flowers, serum, cream, liquid soap, handmade cosmetics. Let's put by, let's click Generate. Let's check some images. As you can see, overall the shape is great. However, the lid is a little bit distorted. I think the cycle one came out the best in terms of the shapes of the product. And you can see the by symbols like the flowers and the wooden ball as well. Now also if you go to my likes gallery here I have the styles that I like and I actually want to work with this image. If I open this an editor here on this image, I actually have a few options here. First, I can download it, Basic download the image. I can make variations, I can upscale it, make it bigger, or I can out paint it. If you remember from Ali, the out painting extends the image. Let's try this first. I'll click on Out Paint here. Let's check it out compared to Dali, where we actually have frames and we need to put frame where we want to make the out paint. Here, the lexica extends all the edges. It does it beautifully. You can see the pattern is extended and repeated. However, I think in Dali, actually, when we were out painting, we were able to specify what we want to out paint. We were writing a prompt, but here, there is no function to write the prompt. It just does it from the style. That's what the out paint does. Also, you cannot choose the dimension in which you want to out paint. If you had the image in a portrait format, you cannot change it to out paint in a landscape format. For example, it will be in the same format as the original image. That has some limitations with out painting here, but the feature is great. Now let's try some other problems. We can use the same prompts as we use for and for images AI. For example, for realistic port, I can use the photograph. For introduction, I said that here we can use very simple prompts for realistic photograph. I want to try that. Instead of basing this whole prompt, I would probably just put a portrait photograph of a young British woman in a jacket. Let's try this. I'll photograph of a young British woman in a jacket with wavy blond hair. I will also put the background blurry, rainy city, street background. We can also add negative prompts, but for now, let's try just this. Here we can see some gorgeous photos of, of a young woman here. Let's say you like this one. You actually can make variations of this image. You've got some small variations. As you can see, her hair are a little bit purplish. In this one you have a little bit different signs. For example, here she's wearing pretty much the same jacket, but as you can see, her hair is merged with the jacket. This is a little bit better here by using variations, you can remove certain artifacts from the image. Okay, now let's try the whole prompt and see if using a longer prompt with Lexica will make any changes. Also, let's add the negative prompt. The images we've got with this longer prompt and negative prompt, I think they came out worse than the images with the simpler prompt. Because look at these images, The background is fabulous. It's like nice, soft and the whole composition is very harmonious. And the light, everything suits very well. Here we have some artificial feeling. I would prefer this image, let's say, because I like this image a lot. I can actually go and out paint. After out painting, I added some more background, it extended. We've got a little bit more of the jacket. This side looks beautiful, but I think the signboard is bit deformed. The sign looks a bit better here. I think the building has some problems here. The umbrella is flying in the air. Creative. I like this one the most. You can download it or make variations of it. You can like it to save it in your likes gallery. Another thing you can also do, you can load prompt into editor. Basically, it will load this prompt here. We can try that load prompt into editor as you can see it out of the prompt here. You can also load image into editor that we'll load this image here. You can generate, you can change the prompt a little bit and generate images similar to this one. But I will show you how to do this a little bit later with our own images. I think this will be more fun. 23. Lexica Image Generation: Let's strike you other prompts. Here we have the logo, as you can see with Lexica, shorter prompts sometimes work even better than the longer prompts. So let's put line logo of a cupcake with chair and top. I want to make it a square, so I'll change the size and I'll click Generate. Okay, here we have some interesting ideas I think so far compared to wit images, This is in a completely different style. Definitely. It looks like a sticker here, so you can see the white outline here. But yeah, here, some images that tried to make it colorful. Did we put colorful here somewhere? No, we haven't. But it was creative here. I want to change it a little bit so we don't get it too colorful. I'll put colorful here. Colorful in the negative, prompt. In the dance settings, I can also choose like guidance scale. That's exactly the same as prompt guidance in images and basically means how close the image follows the prompt here. I wanted to follow more. I'll put maybe nine here. The maximum is 13 that we can choose. I'll put, let's say nine here. Let's generate again. Okay, here, I think it's a little bit better. I don't know how useful that would be for a logo, but it would be great for a menu. I think. An image for a menu or sticker. Here's the style, let's try a different one. Magical realism. We have the three D render of Raccoon. I will delete the Unreal engine from here. And also negative prompt, and we'll make the guidance scale back to seven, the regular one. Let's click January here. I think the raccoon is missing the bottom half. Let's check the other ones. Okay, we have this front of raccoon on the armchair and the bag is from behind. Interesting. I don't see any legs but maybe the hidden in the book. I think this one was the best one, because here at least we see some by clicks here. I'm a little bit disappointed with how it ended up. But let's say if we improve the guidance scale a little bit, maybe also nine, and see if that will help. Okay, this didn't help. It made it even worse. But yeah, it added a little bit of details. In the sofa, you see like a small patterns and so on. Let's without just con reading a book, arm chair, and lamp and see if this will work. Let's make it eight. As you can see, this is actually a little bit better. Maybe the three D render confused it but I don't know, it looked like a tail. But now as I'm looking more on it, it's not a tail. We have some artifacts, that's definitely the problem here. Maybe this model is not that well known animals, I'm not sure. Ok, let's try other ones. Illustration children's book illustration. Let's try the illustration here again. I'll take out the artists names. Girl riding a bike. Okay, let's try that. Let's move it to seven. Okay, this is way better. Legs are deformed here, but everything else looks perfect. The background is very nice as well. Here. Yeah, not bad for illustration. It did a great job. Okay, let's move on to landscape. I paste my prompt here. I wouldn't remove anything from here because I think that's pretty descriptive. I've actually seen long prompts in Lexica. I wouldn't worry too much. I'll make the dimension, maybe this one guidance scale, let's make it smaller. Ai has more room for artistic style here. We've got our magical castle here. Again, I think I did an impressive job with the castle and the landscape. The colors look astonishing. Okay. The landscape, it did. Well, let's share the conceptual art here. The meaning of life. I will keep all of this here. Maybe I'll make it. Yeah. Let's keep the same landscape format here, Guy scale. Let's keep it a little bit lower than 55 would be good. Let's generate, wow. Look at this image, this is amazing. There's some tunnel in another world. And you have a ship, a moon, and that just looks magical. You can see a small house. There's so many details. I love that. Everything else also looks pretty good. And then there's a person standing, observing the beautiful. I don't know, looks too large for anything, but maybe it's a different planet, so who knows? And here we have a rainbow and a galaxy in a wave, I'm not sure, but a beautiful merge and a person again here observing chanting scene around him. So yeah, that's some photos we've got. I'm not sure why we cannot upscale this, but I'll save it. 24. Prompt Guidance Parameter: We've tried all the prompts that I've prepared for you with Lexica Art. Now, I would like to play around more with guidance scale and show you what effect does guidance scale have on images. For that, it's best to have a prompt that has a lot of elements in it. For this one, I chose a prompt with a lot of things going on. It's a girl holding a tiny kitten. A girl holding a tiny kitten in her arms. Waits for a bus at bus station. Here we have a lot of details. We have a girl and she's holding a Keta. The background should be a bus station. Maybe we'll see bus. Let's try guidance scale of middle, maybe a seven. Let's see what it will generate. The images we've got here depict exactly what we wrote in the prompt. So we have a girl and a small kit, and she's holding the key ten. I'm not sure if it's a bus. It looks a little bit more like a train, but it could be here. It's definitely looks like a bus. The kitten is bigger than the girl. Okay, let's count. How many fingers does she have? She has 123456 fingers. That's the problem with AI. It gets the fingers wrong. Maybe we can put that in the negative front, but I'm not sure if that will help. Extra fingers here. I think this is one of my favorite one. She's holding a kitten and I don't know if I tried to make it into a jacket or something, but that just looks like a big blanket with the hand, it got it correctly here. Five fingers here. The proportions are all messed up. This one was the best one. This is what we've got with guidance scale of seven. Let's now do the maximum. Let's do 13, that's the maximum we can put here. Let's generate again on this image we've got some past station which was missing from any of the previous images we've generated here. The egg. Got the fingers right by fingers. The cat looks a little bit bag. What can you do here? The kitten is small. I love this one. Again, problems with fingers guidance scale of 13.7 The images are aligned with whatever we wrote in the prompt. However, I would say that the images with guidance scale of seven, they feel more natural compared to the ones with the guidance scale 13. I'm not sure. Here I find that we have more artifacts not just with hands, but with the jacket. And the whole composition feels a little bit more forced compared to these ones, for example. So this is something you need to be careful with. So when you increase the guidance scale, you may get more artifacts and the composition may look a little bit more forced. Because now AI tries to integrate all the details that we've included in our prompt. Including the bus station. It tried to integrate as much details as possible in one image. Let's now try the guidance scale of the low guide scale. Let's two and see how that compare to everything else. Again, we have a girl holding tiny kitten and negative prompt with extra fingers. Here we've got the images way more darker. If we compare these images to the ones we did with the guidance scale of 13 or number seven here, it feels more cheerful, bright, and even though we didn't specify any lighting here, but with the guidance scale of two, the colors feel dull and also the whole atmosphere feels gloomy. I want to emphasize this point when you increase. Guidance scale, the contrast and color saturation will increase with the higher guidance scale number. If it's a lower number, then you would get more foggy and less saturated colors as we can see in these images. Also, I want to focus your attention that in these images, in some cases, we've got exactly what we asked in the prompt. So we have a girl here, she's holding a kitten. And that does look like a bus station. However, on others here, this is not a kitten, this is some other animal. And here the background is not clear that it is a bus station here as well. When your prompt guidance is on the lower side, then some elements of your prompt may not be reflected in the images and that's something that you need to be aware of. If we go to stable diffusion guide, this is a guide to guidance scale parameter. Here we have a panda playing a guitar. Here's the guidance scale of eight and this is the image that it has. If we make it small, let's the smallest one. Guidance scale of one. Here we've got something quite random. It doesn't look like a panda anymore. But when we increase this, now our image starts to look like Panda playing a guitar. Again, look at the colors here we still have some foggy, more unsaturated colors. As we increase, we've got more contrast and more saturation going on. I think guidance scale of ten is pretty good here. Number 12 works. We get a little bit more details here now. Let's 17, 18. I would say 13-18 It feels quite similar. But then let's zoom in a little bit so you can see better at number 20. We are here. It was to move here. Now if you look, we're starting to get more artifacts. Look at the guitar, look at the eyes. So the whole image starts to look worse. As we move the guidance scale even further, you can see the image is deteriorating at the guidance scale 30. Here we have oversaturated image, the quality is very poor. You can see that the whole image is pixelated and the quality has deteriorated a lot. Here we can read that the most creative and artistic results are usually generated around a guidance skill of seven. But using a skill up to 20 still produced results with little to no artifacts here. For this image, the best guidance skill value is, in my opinion, between number 8.18 and then just the quality is getting worse. But it all depends on your image and what you're trying to achieve. If your prompt is longer with many elements at it, maybe it's worth trying a little bit higher guidance scale to make sure that the image incorporates all those elements. But sometimes it's worth trying to do a smaller guidance scale if you're doing more abstract art. So it really depends on your artistic vision here. For guidance scale, that's all that I wanted to show you. In the next video, I would like to go over how you can upload your image and how to do image to image generation with Lexica. See you in the next video. 25. Lexica Image to Image Generation: In this video, I want to finally show you how you can do image to image generations with Lexica. It's very simple. All you need to is click on this button, Upload Image Here. You can choose any image from your computer, or if you found image from Lexica, you can click on this button and click Load Image into Editor. And that will load the image here, but for now I want to use my own image. So I will click Pod Image and choose this ballerina that we used with Ali. As you remember, that was a catastrophe with Ali. Once you have the image here, you need to write a prop. Basically, you should describe your subject and what you want to generate here, I want a ballerina dancing. Instead of this white background, I want a magical forest. A ballerina dancing in a forest. Okay, let's try this. In these images, we can see that the background is more or less that we have in our source image, which is pretty much white here. It got the hands wrong. Here we are getting a few more elements in the background. Here we get butterflies and a few stories. This is a little bit better, but I want more of the force. I want to see trees, I want to see leaves and so on. This is what I'll put trees and leaves. To make sure that we do not get this white background, I will put white background here. Let's try this. Okay, this is a little bit better. We get some problems with still with the leg here, but the background is a little bit more detailed. Okay, here we've got quite good legs. Okay, to improve this. To improve the background details, I can use the guidance scale. And I can make it, instead of seven, I will make it ten. Because now it's forced to use the word. It's forced to have magical forests in the background. As you can see, for dimensions, we cannot change the dimensions. The image that will be created will be the same dimension as our original image. Okay, let's try the guidance scale of ten. As you can see, increasing the guidance scale here actually improve the background a lot. Now we have way more elements of the forests in the background. This one looks pretty. Let's try one more time and let's make it 12. I also want to put fantasy fantasy. In order to avoid poorly drawn legs. I will also add extra limbs. I'll put extra in the negative, prompt. Extra extra hands. I have white background, extra limbs, extra legs, extra hands, Poorly drawn feet, poorly drawn face. This is something that I don't want to see here. Okay, here are some images, let's check them out. The legs are a little bit better here, but we do not get the back around three legs. I think this is the best image so far in terms of the legs, the hands, and the correct facial features here, the posture is quite similar to the original image we have. Here. We have this beautiful, magical forest background. Okay. As you can see, she's standing on some, a lake that's beautiful. Now, I would like to explain a little bit, how does AI generate these images from our image? Basically, the generator doesn't use a single pixel from this image. What it does, it analyzes this image and then converts it into code. It then uses this code. Input to generate all other images, you won't be able to get exactly the same image, but you'll only get the variation. As you can see here, the posture looks quite similar, but not exactly the same one. Again, I will try to capture as much detail from your original image and integrated in the new images. But again, some details or composition may be quite different. Or it may not capture, for example, facial expression. Or just facial features may be quite different because it may not capture well the information from the source image. Now I would like to create image to image generation with my own photo. Let's try that. This is my photo myself. Here I will put, I'll describe myself. A girl with curly hair. Now I want to make images in anime style. Yeah, I'll keep this. And then maybe in the CT Street background, let's make the scale maybe ten. Let's see here, the AA actually captured my blue and white striped dress quite well. The overall posture of the girls is similar to my image here. However, none of the girls here look like me. The reason is that we've provided AI only one image for certain things like posture colors. It's certainly easier to give a description when it encoded. It's easier compared to facial features. With facial features, it needs a little bit more extra information. If it had more images of me, then it would be easier to compare and see what are my facial features. However, here, because I provide only one photo, there is not much I can expect from AI. Let's try to generate a few more and maybe I'll change the guidance scale to seven, back to seven, let's see. Okay, as you can see, it captured the wavy hair, but everything else, again, the face is very different as you can see. If you were to upload a photo of yourself, you would get a completely different person. But the posture, the colors we built, the clothing items look quite similar. I would recommend using maybe a full body images, because image to image generation does the posture quite well. And here you can experiment and try all different backgrounds. You can get really creative here compared to Dali. These are way better. That's it for Lexica. In the next module, we will cover more AI image generators. See you soon. 26. DreamStudio.ai Introduction: Hello everyone. In this module we will cover another AA image generator. But before I begin, I wanted to cover a little bit more about stable diffusion because I feel like that we didn't get a chance to properly introduce it. Stable diffusion is a deep learning text to image diffusion model, and you might be wondering who developed it. It was developed by the start up company called Stability AI in collaboration with academic researchers and nonprofit organizations. One of the collaborators is Runway ML. This is actually an AI platform right now for AI image generations, for AA editing, and for video editing and video generations. We will be covering Runway ML in our course as well. Stable diffusion was released pretty recently in August 2022. It's open source model compared to Dali and Mid Journey that have their models closed source. That means nobody can access them. Stable, stability. I actually made their model open source. That means everybody can access it and use it as they wish. It has free license for commercial and noncommercial use. Because of that, you can actually write it on your personal computer for deli and mid journey. Of course, you get some free credits. But after those free credits, if you want to generate more images, you have to pay. But here you can, you can use table diffusion as you want, and if you run it on your personal computer, you don't have to pay anything. It's free. You can generate as many images as you want. So that's the beauty of the open source model, stability. Now that you know that table diffusion was developed by Stability, I, I want to show you their website. This is stability website. And they actually have a few products here. One of them is Dream Studio. And Dream Studio is an AI image generator similar to images. However, here you actually need to pay for image generations. You may be wondering if stable diffusion is open source. Why should they pay for Dream Studio? That's basically for example, for some reason you cannot run stable diffusion on your computer. Such as if your computer has low compute power, then you can use stability is compute power to generate your images. You'll use Dream Studio in that case. They also have other great products. Clip drop, we will also cover that in our course. That's for image editing. There's also Photoshop pin and blender pin. We were not covering pins in the scores. Okay, that's for stability for Dream Studio. This is what will be covered in this module. It's an image generator, it's stable diffusion. It's a web app hosted by stability. Let's go to Dream Studio. This is basically studio interface here. Now let's talk about some advantages. They give you free limited credits, which is great. And after that, you actually need to pay based on your usage. They don't have the subscription, it's based on the usage. You can generate images in different styles. As with any stable diffusion, the images can be generated in various styles. It has advanced settings. If we go here, you can change the dimensions of the image. You can change how many image you want to see generated, as well as you give a prompt guidance. Prompt strength here is the same as prompt guidance. You can put generation steps and seed number. We'll talk about seed later as well. You can choose the model here, okay. It also has a image editor. If we go back here here, you can click on the edit and you will be able to upload your image here. And we'll do in painting and out painting as well. In terms of disadvantages, Dream Studio doesn't have user friendly interface. Actually, I was surprised because Dream Studio is a product of stability AI, which is one of the leading companies in AI and all the other products are quite good. But the Dream Studio, I found that it can be a little bit buggy and just the whole interface doesn't feel that good. Another problem is that it's beginner friendly. If you want to use some advanced settings here, you actually need to put everything here yourself. You have to know exactly all the terminology here. Compared to, for example, images I, where they have lots of images where you can choose like styles. They give you hints in terms of which prompt guidance to choose or like steps, for example. They use like draft or detailed words to help guide you. Here you have to know everything yourself. But at this stage, I think we've covered this terminology. You should be pretty good. Another thing is that it requires detailed prompts, because Dream Studio is basically stable diffusion, as we talked in our prompt writing module. In order to achieve good results, good image results, you actually need long prompt. Here you go. Great way to practice your prompt here. Another disadvantage is that it has only a few stable diffusion models here. Put SD, that's an abbreviation of stable diffusion. If we go here here we only have the three new stable diffusion model. It's stable diffusion version 2.12 0.1 768 and the better trial model of SDx L. As you remember from images I, there are a lot of stable diffusion, fine tuned model and also their previous base models like 1.5 and so on. Actually, I personally prefer working with other stable diffusion models. This is a big disadvantage for me, that here I can only choose these three. That's it for Dream Studio. In the next video, we're going to go and explore some features here and try out some prompts with this Dream Studio. See you soon. 27. DreamStudio Features and Models: Now I would like to show you Dream Studio first when you sign up. All again, this is the interface that you will see if you click on this generate pattern. Here are a few parameters that will go over. The first one is style and what style you want your image in. Here are some options. For example, anime comic book, digital art, fantasy art, Neon Punk. Then we have some isometric low poly origami line art craft, clay, cinematic D model in pixel art. If you want to generate in one of these styles, choose the style. Otherwise you can just keep the default one here. We can write a prompt here, you can randomize, it will just give you a random prompt here. In the negative prompt, you can write some negative prompt, something that you don't want to see in the image. Then you can upload an image if you want to do image to image generation similar to what we did with Lexica. Another thing here, if you go to Settings here, you can change the dimension of the image. This will be vertical, If you go here, it will be horizontal. Here you choose how many images you want to see generated every time you click Dream. By default it's four. But if you want to save some credits, you can put maybe two or even one. You can see here, if you want to generate this image that's horizontal with one image count, that's 2.6 credits. If we go to advance, here's our width and height, that's proportional to our dimensions here. Then we have prompt strength, which is the same as the prompt guidance. We have the generation steps. We also have a suit and we will talk about sit a little bit later. Okay, as you can see, as you change the dimension, the number of credits also changes. If you choose some, either horizontal or very vertical, it will be the highest number of credits. But if you choose a square, it's the cheapest number of credits for model. Let's login for model. There are three models, these are all most up to date models. Stable diffusion version 2.1 is up to date. Stable diffusion model available publicly. There is another one called SDL and that's just in a better mode. It's not public yet, so you can only try it in drip studio. This is something that I wanted to talk about here because models are updated regularly. In a few months, you'll likely see maybe some different models. That's why it's important to check out what is the model about what kind of prompt you need to use for that model and so on. So for example, for stable diffusion version 2.1 from the sources, I've read that the negative prompt for this model is super important, which may not be for some other stable diffusion model. It's always good before starting to use any model to read a little bit about the model. Here are some articles about the stable diffusion version. 0.2 0.1 by stability I, yeah, they describe a model a little bit. They say what they added, how it works, and so on. Another source that I really like for checking out the different models is the stable diffusion. And they have great guides here. For example, for 2.1 model, for example, as you can see here in Dream Studio, you have the version 2.1 and version 2.1 768. If you're wondering what is the difference, if we go here and here, it says that there are two text to image models available. 2.1 base model, which has default image size of 512, 512 pixels. Or. The 2.1 model, 768, which has the default image size of 768 by 768 pixels. The 768 model is capable of generating larger images. It's especially useful for generating larger scenes with small characters. Here's some description of these models. Now we have this SDX. Again, you can go to the source Stability AI. Let's see what they say about the model. Highlights of SD L capabilities include next level photo realism capabilities, enhanced image composition and phase generation, reach visuals and jaw dropping aesthetics, use of shorter prompts to create descriptive imagery and create a capability to produce legible text. From all of these, I would say the most important one is that this model can produce legible text because all other models were not good with text. We can actually try it out here. If we go back to Dream Studio, let's choose the SD Cel beta, Let's keep the square, and I'll choose image count. Yeah, let's do four here. Now, I'll choose a prompt that has some text in it. The style. Yeah, let's do enhance here. The first prompt we can try is a photo of a man holding a sign that says thank you and I want it highly detailed. Okay. As you can see here, the I got thank sign very well legible, no artifacts create. So let's compare that to the previous model. If we choose stable diffusion 2.1 again, let's generate it. As you can see, there is a huge difference even though it tried to write. Thank you. But all of these are just artifacts. Now we know that this model is great with text. Let's try something more difficult. Photo of a bus stop advertisement displaying a burger in a text, Hungry Close a View, highly detailed, 64 K. Let's try that. Let's do the Dl beta model. Okay? As you can see here, the first image is missing the hungry sign. However, the other three you can see here clearly, the hungry and the burger. Here are a few artifacts, but this one is one of the best ones. I would say maybe we can try one more time and see if we can generate better images. Okay, here it actually disregarded the message completely out of these images. I think this one is the best. And I would probably change it a little bit. I will change the prom strength to ten. I will also display an image of a burger and a text, Hungry. Hopefully that will help. Let's check it out here. I think only the first image depicts my prompt correctly, although I don't like that. It's black and white. Here's actually, you have an editor. You can go to Editor here. You can click Edit Image on the right hand side, you'll have the image. You'll have frames. You can add as many frames as you want basically. But for now, let's remove all the frames. Let's add a new one. Let's say I don't want it black and white. I will erase this whole advertisement here, and I want it to make colorful. Let's do that. Now, since we have this frame here, it captures it. I found a little bit hard to work with editor here because it's baggy. For example, here I cannot move the frame. Sometimes I cannot move the frame. I usually need to restart it. Let's try to restart it. When I restarted, my image disappears. Let's go back here. You can click to edit this image. Finally, we can move our frame around again, I have to erase this image. Now, I'll move my frame to the place where I want generation to happen, which is here. I will line it with my image. Now I will write a prompt. I will use the same prompt. I'll put an advertisement, tisementoardsplaying hamburger and text hungry. I'll put that highly detailed. In the negative prompt, I'll put black and white image count. Let's try, let's put prom strength at 12 to make sure it aligns with our prompt better. Let's see. Okay, not bad. Yeah, it edited the part that I've raised. It didn't add the text here, but on the next one it did put the burger and the text exactly as I wanted. This is the tool that you can use as well. The interface is not great. Just heads up. 28. DreamStudio Image Generation & Seed Parameter: Okay, let's go back to generate a few more image. S Let's see what are some other improvements of the model. If we go to Stable diffusion Art guide about the SD Excel model, here we have a person writing their own experience about this model. For example, let's go to Improvements, legible text then. It's better human anatomy, the postures, we'll try that as well as you can see the differences between a yoga practitioner here and the images with the previous stable diffusion model and more aesthetic images. Here we have a house, and here's our indoor setting, as well as the style you can see is a little bit different in the images. More accurate images. The ability to understand the prompt improves over version on E models. Here we can see or tone portrait of a woman here in the previous version we got the black and white, and in the newer version it actually used a variety of colors. Let's try something else for Dream Studio, we've got this burger and now I fought off a modern bakery with minimalistic interior design, with clean lines. History showed in glass, displayed contemporary environment, highly detailed. I want to make sure that the bakery has a sign with the text bakery displayed on the wall. Let's paste this prompt and try it out with the new version. For the negative prompt, I will put poor proportions blurry. I don't want it to be unclear. Okay, let's try that. Okay, on the first image, we do not get anything here. Looks like bakery, but there are too many artifacts. I'll try again. Every time you make a new generation, your advanced settings reset. Make sure you change the advanced settings before you make a new generation. For prom strength, I'll put 12 here. I will use Dream again. As you can see here, I still try to incorporate bakery. Here we have two Science, the second one is more legible. However, my takeaway is that this model still struggles with text, especially when it's a little bit more complicated here, because it has to be three D and it has to have the right proportions Compare to, for example, the first images that we generated where it's nice and flat here, Here it did a great job. Still needs some improvement in this area. Let's strike one more prompt. In the article that we've read, it says that this model is way better with postures and I want to check it out. I designed a prompt, a wide overhead, short yoga practitioner in a tree pose, mountain setting, a soft morning light by Thomas Moran, Highly detailed, let's try that. In the negative prompt, I'll also base my negative prompt, bad framing, out of frame, deformed, and so on. Make sure to choose the style. For example, here I wanted a little bit more photo realistic. Actually, I'll choose the cinematic in the settings. I will make the prompt strength even further. Maybe 14 generation steps. I'll keep that the same. And the newest model, let's try that. Okay here. Actually it does somehow looks like a tree pose. If you're not sure what is the yoga tree pose? Tree pose. Yoga. That's how it should look like. As you can see, the images that were generated have this exact pose. We can use the other model, the 2.1 version, and see how this model compares to the S D L one. Here with the 2.1 model, we got this image which is not bad. This one? Yeah. Here, the proportions are messed up. Yeah, I would say that the newer version has better proportions. Okay. For one last thing, I want to explain to you what is seed, so you can already start using it. A Ed is a randomly generated number assigned to an image. Every time you generate an image, it will have a different number. For example, this image here has a number. This one. This number tells AI how to generate the image. What it's great for is that if you use the same prompt and you use the same settings, the seed number, you'll get exactly the same image. What's even better is that you can make small changes to the prompt by using the same seed, you'll get almost same image with slight variations. Basically, you can make small variations to the image. That's very important for artists. Let me show you what I mean if we go here. This is, by the way, an article about the seeds, but it has great examples. I wanted to show you that as well here. This is the first prompt when the person generated this image. This was the number when they use the same prompt and settings and added this seed number. Also, they added smiling to their prompt. Here they've got the same girl, but now her mouth corners are left up. Then instead of smiling, they'd be added angry. And now you can see also basically the same girl, but now her expression seems like she's angry and here's excited. The same thing you can do with the landscape. This was the prompt of a park and they used the same seeds and they only changed the time of a year. Here's the spring, now it's summer, autumn, winter. As you can see here, the composition is exactly the same. It's just the color of the trees is different. Here's another example of Elon musk. Again, same composition. However, here they changed the medium. Now here it's by Vincent Bango, here by Pablo Picasso, Salvador Dali, and so on. This is how you can modify or improve your image. Let's try that. Let's go back to Dream Studio to try out how to see work. I've prepared another prompt, and that's a portrait of a young woman with a Asian market background. For the negative prompt, I'll add the basic negative prompt here. Now let's choose the style. I want it to be cinematic advanced. Make sure the DX model is selected. Let's make the prom strength. Let's put ten. Right now, we don't put anything for the set. We first need to generate something. Let's try that. The images I've got, I don't quite like any of them. I think some of them have artifacts and the other just the face is too dark. I will change the style from cinematic to enhance and try again. Okay, here, it's way better. It's either this image. Or this one. Okay. I'll choose this image to work with. Here you can see it's seed number. Let's now we can paste it. Let's paste our seed here. Now let me show you that you can actually generate the same image. I will make the image count to one using the same settings. We will generate our prompt with the seed number. You can see here we've got exactly the same image as before. Exactly the same. Now what I can do is to add a few modifications to the prompt. They should not be big modifications because with big modifications, the whole image will be completely different. But I want to keep my subject the same. Now I'll add smiling a portrait of a young woman smiling. Let's try that, make sure the seed is the same one. Let's try that. In this image, you can see that the composition is almost the same as the image before. And if we zoom in a little bit, so this is the smiling one, you can see her math corners are lifted up here. Yeah. Basically we have the same person here, the same subject, the same hair and background. However. And a add, try to add the hand. Not successfully, unfortunately. Okay, now we've got the smiling one. Let's try frowning. Here, you can see that we've got the same person here. Same hair, clothing item, and facial features, as well as the background is quite similar, the composition is different. This is the front view. Here we have the side view. You definitely can see different expressions. That's the beauty of using the set. Now you can use the set to generate a character and make different images of that character with different emotions or different postures. You can play around with the set. Now you know all those advanced parameters that dream studio have. Because we've already talked about the prompt guidance, the generation steps, and now we've also talked about the set should be all set and try it out. Personally, I don't usually use Dream Studio because I'm using the stable diffusion on my computer, which is free. It also gives me a little bit more freedom because I can use any model that I want. And there are more advanced settings, but Dream Studio has its own advantage, is that this newest model is the Excel is not released yet and you can only try it with Dream Studio. So here you'll be able to try any new models that stability AI is planning to release. So try it and see you in the next module. 29. BlueWillow Introduction: Hello everyone. In this module, we will talk about another AA image generator, and it's called Blue Willow. If I go to gallery here, here are some images that were generated using Blue Willow. As you can see, there are some high quality images. Okay. Blue Willow was founded by a group of AA engineers. It was launched in January 2023, and it operates on Discord. If you haven't heard about Discord, Discord is a messaging platform. I would say quite similar to telegram. For you to be able to use Blue Willow, you need to have an account with Discord. But don't worry, I'll show you how you can set it up. What is unique about blue willow is that it's an aggregator of multiple AA models, including models like stable diffusion. What it does, it picks the best model based on your propped. For example, if you write a cartoon image of a dog for example. It will choose a model that's best for cartoon style images. That's basically what it will do. So if I go to their questions and answers here, what makes it unique from other AA text to image generators? Blue Willow is like a Google flights for AI models. It enables users to find the right model depending on their goals. Unlike other AI text to image generators, Blue Willow is an aggregator of multiple AA models, including models like Stable Diffusion. Who owns the rights to images produced by Blue Willow? You own the rights to your creations. You're free to use them in your art for commercial gain. Here's some information and you can further read about blue willow here. Okay. What are some advantages for beginners? It has some free limited generations. Currently, it gives ten generations per day, but of course you can buy a subscription with higher number of generations. It's beginner friendly if you have a Discord account. Let me show you here. All you need to do here is go and click Imagine, Write your prompt, and it will generate images for you. You can upscale, create variations, and do out painting with your images. You can also do image to image generation. This is not trivial how to do it in discord, but don't worry. I'll also show you how to do that for disadvantages that it needs a Discord account. It's not like a website where you can go and try it out. No, you need a Discord account. There are limited settings here, even though it's stable diffusion. For now, Blue willow doesn't have any way where you can add the seed number or alter the prompt guidance steps, number of steps. Or choose your own model. Because it chooses the model in the settings for your image based on your prompt that's handled automatically. Also, of course, because the models they use are stable diffusion models. So detailed prompts work better with stable diffusion, and that's the blue willow. In the next video, I will show you how you can create a discord account and how you can add the blue willow there. And we'll go from there on. 30. BlueWillow Overview and Discord Setup: Here's where we left off in our previous video. Now I want to try some prompts I get. Let's try our props, the realistic photo, and so on. Now, I actually made some modifications. It's still a photo, but I've changed the British woman to Indian woman and also changed the background. I modified my prompts a little bit so we get a little bit more different spectrum of images here. Okay, let's do that here again. You need to put, you can click on the skin here, or you can type whatever you want. Imagine, and then you need to click space. Now we can it be our prompt here. Here I have the professional portrait photograph of a young Indian woman with long hair, beautiful symmetric face. Cute natural make up colorful street market background. Highly detailed sharp focus, deba field, and aperture. And okay, let's try that. Just click Enter. We will get our results promptly. Okay, let's check this out. Okay, as you can see here, the phase looks good. The eyes are not messed up, the nose is not messed up. Facial features are correct, which is great. Here we have the four images here. Let's say if you like any of them, you can upscale it. For example, the first looks good. I can go ahead and upscale U One, It's short for upscale. That refers to the first image. This is the second image, will be upscale two. This one is the third image, three, and this one is the fourth image, four. Let's upscale the first image here. Here's our image. The eyes are not that great as we look on the upscaled version. These buttons you can use if you want to out paint this image. If you want to out paint to the left, you can press this left arrow if you want it to the right, then right arrow up and then or down. If you want out painting in all the directions, you can press this button here in the bottom, we have Mogi also cross. If you don't want the image to appear here, you can click this Cross button and it will disappear. Also, as we've talked about, blue willow has feedback, don't like the image. Then you can click on the Emoji. Or if you love the image, you can also give them feedback. Actually, they also right here, rate your image. After upscaling your image, you'll see new emoji buttons that allow you to rate the image from worst to awesome. This helps us a lot in improving our trading data. You can rate your image and help Blue Willow improve here. For example, I would say that it's okay. I'll put maybe this emoji here also. I will click out painting so you can see what that does. Okay, let's go here. This is our out painted image as you can see it out painted in all the directions here. And again, you can choose the image that you like the most. Go ahead and upscale it. Okay. Now I also want to talk about parameters because as you can see, you can just put, imagine, then it's only your prompt here, where can you put negative prompt or how you can change the dimensions of the image? All of this are in parameters. If we go to blue willow dogs, they should be here are blue willow dogs. Here you can see this prompt and parameters here, all the parameters that blue willow has. For example, they have a negative command. This negative command is basically a negative prompt. Now you can imagine. Here's your prompt painting of a cute cat. Then if you want to put a negative prompt, then you put two dashes. Then no, you put anything that you do not want to see in your image. For example, here you don't want to see the three D or cartoon. Let's try this out with some simpler prompt here. For example, I have magical realism. As you remember, we've had the **** reading a bog. Now I've changed it to three D Render of a panda playing chess with a rabbit in the campy home. Dim, lighting realistic, unreal engine. If I go back and paste it in my imagine prompt again, I will paste my prompt here. Now all I need to do is to put two dashes, no, then something that I do not want to see here. This is a three D render. I do not want to see any cartoon. I also don't want to see any extra legs or extra arms. That should do now. We just click enter and see. Okay, let's see if I zoom in here. The panda has rabbit ears. For some reason, I don't like that here. This is a little bit better. Again, rabbit panda with rabbit ears. Playing chess with a human. Again, panda with rabbit ears. Okay, that's not too good. Then let's try something different. Let's add that in the negative prompt. Here I want to put, imagine the same prompt. Now I don't want cartoon extra legs, extra arms. I also don't want to see Panda with rabbit ear, rare bits ears. Panda with the rabbit ears. Okay. Hopefully that will fix it. Okay, Let's see, the first one, we got some animals. Actually it does look like a panda and a rabbit. Again, I'm not sure if this is a panda here, but on the third image, I think that looks good, actually. Again, you can upscale the image that you like or you can make variations of the image. Again, the number corresponds to the image number will be this, one will be 23.4 because I want to make variations of the image number three, then I will click on three here. Also, while it's rendering, this button is also the same cross. If you don't want to see these results, you just click on this button. Or if you want to redo this prompt, you can click on this button and that will just redo and give you more images for the same prompt. Let's see what we've got. Here are some variations of the images. Here are some different postures. Again, we are getting a little bit of rabbit ears. But I think number two, actually 12 or four works for number four. You can see that the rabbit has double ear here. Let's try to upscale it and see if that will fix it. Also, if you've noticed, there is an image of the rabbit which is quite neat. Here we have some human portraits, but here is a rabbit. Okay, let's upscale the number four. As you can see here, the upscaled version didn't remove the defect with the ear. Sometimes the upscale helps to remove certain defects here. It didn't just to let you know that whatever you'll see in the small image will be in the upscaled version, Okay, now again, you can rate this image, let's say not good because of the problems with the ear. 31. BlueWillow Image Generation Part 1: In this view, I want to introduce you to Discord. And again, it's a platform for messaging and it's widely popular for programmers and crypto community. And now it's also becoming very popular for a community. If you're wondering why Blue Willow is on Discord, they actually have an answer to that. So why does it operate on Discord? Discord is a community platform that allows members to share and discuss the images they're creating, as well as participate in contests, discussions, rewards, and events. Discord also enables blue willow to gather feedback and improve the platform quickly. We plan to lodge our service outside of Discord soon, so stay tuned. Discord is this community platform where you can do a lot of things. How does it work? Basically, you have two options here. You can download Discord on your computer, or you can use it in the browser. I will use it in the browser. I will click on Open Discord in browser Here, I'm already logged in, but I will log out. If you don't have a Discord account, then you'll need to register. You will need to click this Register button and put your e mail address, create a username and password, also the date of birth, and they also ask for phone number verification. Okay. After that you can login. So let me log in. After you login, you will have something like this on the left hand panel. I have a lot of servers here. But in order to add Blue Willow to your Discord account, all you need to do is to go to the Blue Willow website. Here they have this Join the Free Better button. Just click on it will take you to Willow Discord. And you just need to accept the invite. Then it's asking if I want to open the Discord on my computer, I'll call counsel and I'll continue on Discord. Now after this, you should see blue willow on your left panel. Here is their logo and this is what you will see. I know it looks intimidating. There are a lot of things going on, but don't worry, we will go one by one. Here is the blue willow server. Here's some information. Let's start here. Getting started. Here's the information about blue willow. You can read it in different languages if you want. The here are questions and answers rules. You can sign up to their newsletter gallery. Okay, now we're going to more interesting stuff. Here are a few chat groups where you can generate your art. If you go to any of these ones, let's say maybe number 23. It doesn't matter which one here. Now put dash and write. Imagine then space and radio prompt here, for example, a t in a box. Now you just click Enter. That's all. Now you will see your image is being generated because there are other people using the same group chat. You will also get a lot of other art. It's easy to lose your prompt. Here we go. This is our prompt, So we have a cat in the blocks. As I was saying that, it's quite easy to lose your prompt because every second someone else is using and generating their own art. For that reason, I always recommend to use the direct message in order to do that. Once you see the blue willow here, just click on it here, you have an option to add it to server here. More experience, you can do that. Otherwise, you can just message the blue willow. Let's put Hello. Now this is the direct message and now we have blue willow in our direct messages. Now I can use the same imagine prompt and I can put a cat in a park. And then I can enter here. Only my images will be visible even though the images I generate are public. However, here at, I don't lose them and it's in one place. Okay? Here. In order to get to direct messages, you just need to click on this Discord logo and you'll be brought to direct messages here. And this is the Blue Willow. Chat here. Okay. Now, I recommend going back to Blue Willow and I wanted to show a few more things that they have. They do have some support here. People can ask questions. Then we have announcements. These are announcements by Blue Willow if something is changing or for example, if prices are changing and so on. Then we have the prompt questions and answers. Which I think is great because for example, if you're looking for coloring pages, you want to create color pages. Here. Some tips and tricks. How to get what you want. For example, background removers or how to generate this one letters and images. Here people are writing the tips on how to create this image with text. For example, here you can see legible text. And then we here you can give feedback to blue willow, you can connect with others. There's also daily contest, this is the description of the contest. Then there are daily themes and you can take part of it and enter the contest in terms of this blue willow server. I think I've wet through a little bit here. Let's go back to the direct messages and let's start exploring blue willow and what kind of things does it have? Blue willow is a bot, Any interaction with the bot will require a slash. I think I previously misspoke it. It's not a dash, it's going to be a slash. Now, when we put a slash here, some commands that we can use with blue willow. The first one is, imagine this is when you want to generate art. Then we have info if we want to read more information about our own accounts. Here I have the information, my username plan I am at how many prompts remaining? I have seven remaining prompts, and this is time to reset these prompts. Then I can go to subscribe. Let's say I want to buy the subscription. Let's go ahead and subscribe. Now I can choose my subscription plan. We have this $510.20 dollar. The $5 gives me early access to version 450 prompts per day, five concurrent images and member badge. The 101 gives me 100 points per day and like member batch exclusive access to VAP contents and so on. For now I'm going to go with the $5 per month Willow. Let's go on here, you'll see a typical payment information. Just add your payment information and subscribe. Now I have subscribed to Blue Willow. And in the next video, we can go ahead and try out some images. 32. BlueWillow Image Generation Part 2: Now let's talk about a different parameter. If we go back here. You can use aspect ratio. If you want a horizontal or vertical image, you can put the and then R. Then it's either three, column two, or two column three. If it's a landscapes three column two, or if it's a portrait that it's two column three. If there are no aspect ratio, then it's going to generate a square here. Let's try with another prompt. For example, here I have anime. And it's a portrait of a skin me boy classes listening to music in the street of a rural Japanese city. Anime boy, high detailed, a sunset, relaxed pink and purple cloud stars, soft light realistic eight K. Unreal engine. I'll change the Japanese city. Okay, here, let's again put the slash, Imagine, I'll paste my prompt here and change the Japanese city here. Japanese. Okay, now we can put no limbs, extra arms. Make sure that you put the no at the very end of your prompt. Because if you put the no somewhere at the beginning, that it will treat everything as your negative prompt. Sure, it's at the end. Now we can also add another parameter. Because it's another parameter, it's fine. It's not going to treat it as the negative prompt. Here we put a R, Let's say we wanted a portrait, so we'll do two, column three, Let's generate that. As you can see here, all the images are in the portrait aspect ratio. That's what we've asked. We have this dime boy here and right now the generation uses version three. We've got some advertisement here as you can see that this is the version three. The version four is the improved version. In order to be able to use the version four, you need to be subscribed, otherwise, it will automatically use the version three. Here you can choose different models. For example, if you want to generate with the first model, you can put V and then space one. Here it will be. Imagine watercolor painting of a cat. Then at the end, you put the space one. If you want to use the second model, that it will be number two. If it's a third model, then this is a default model. You don't need to put anything. If you want to use the four version, the newest version, then you need to put the four and it's only available for subscribers. This is what we will do now. We'll go back to our discord. I will copy this whole prompt here. Just copy it here again. I'll imagine I'll paste my prompt at the end, I'll put version space, and I'll put number four here. Then I'll click Enter. As you can see, these images are way better, especially I like the number one, Even the number two, and the number four looks very realistic. Let me upscale the number one. I'll just use the one here. The color palette is amazing. I love this purple pink sky and how it matches with his hootie here. Overall, I think this is amazing if this is version four. If we compare this to the images we got with the version three, these look way more simplistic. Okay, let's try a few more images. Again. I'll imagine for the next prompt I have the landscape and this is something I took from prompt hero. I've changed the prompt a little bit here, but the images at prompt hero were super good. I want to try this out and see if it's going to be good as well with blue willow. For now I'm not going to put any negative prompt, but I will put the aspect ratio again. I'll use the portrait one and then we will try also the landscape, 223. Let's do version four. Version four, the images we've got look amazing. The rock, I don't know, in the small river looks interesting here we even have the small waterfall. My prompt was actually the Lost Valley rock arch vegetation, exotic forest and plants landscape concept art. And then there are a lot of stylists as well. Here we can zoom in and try to see which image is better. I think I would choose number two. I can either create the variations or scale. Let's upscale it. It's image number two. Okay, looks impressive. Let's try to out paint it. We can out paint the whole thing or into a specific direction. For example, let's try a specific dimension. First, I want to extend my image to the left. Let's click on the left arrow. As you can see here, we've got four different variations in the left panel. It added this little section here. It added different elements. For example, here, it actually tried to add a completely different image and it even put the line here. On the other ones, it looks more realistic. For example, number 3.4 looks very natural here. We can keep any one of them. Let's, for example do number three. Now this is the app scale version. As you can see here, we cannot actually do more out painting. We can only rate the image. I think it looks nice. So I'll put this loving mog here, since we cannot out paint the image further. Which I think is a pity because it would be nice to have the image extended to the right as well. We'll just have to leave the image as is. Okay, let's finish with our prompt. I have a logo here. I've also changed the logo a little bit. It's a tree inside, a water droplet, slick and nalmalistic logo. To graphics, we color white background, contemporary style, perfect for a modern eco friendly business. Tailed eight K. Let's imagine and I don't want to see any three D image, I'll put three D, the aspect ratio I want to square, so I'm not going to be adding anything here. I wanted to be a version four. I'll put version four. Let's see these images. I think for our prompt, it done a pretty good job because this one for logo, it's a bit more complicated compared to our cupcake with cherry Here, it's a tree inside a water droplet. As you can see, we have this tree inside Water draw pled. I like the reflection here. I think for logo, the best one would be more simple ones, probably number three. However, we would need to move this reflection down here. Definitely for logo, we would need to work more for images. We cannot just use it straight away from here, actually. Now I'm curious how. Would blue willow picture our cupcake with cherry on top? So let's try that one as well. Imagine as you can see here for the first and the second one, we didn't quite get the line logo of a cupcake. Probably Blue Willow chose the wrong model for us. It looks like cartoon style cupcake, but number 3.4 looks good to me still. I wouldn't say it's a line logo, just has too many colors. Maybe it can be used for a menu or a website, but not for a logo. Okay, we have one more prompt. That's the conceptual art, the meaning of life. Let's see how blue willow will work with that. Let's make it a landscape. I'll put A R and then three, column two. I'll also put the version four. This looks fabulous, even without upscaling. Let's zoom in. Yeah, all those details. I love the color choices here. Here we have a lot of details. Beautiful landscape here, the trees amazing here. We even have this, I don't know, a town or a futuristic city. Let's choose the best one. Oh, here, it's actually a tree house, that's fun. On the third one, we have a few people walking or moving towards the sun. Let's make the third one bigger. I'll scale it number three. Here we have this beautiful light and the reflection of it in the river. We have these people walking and look at those trees, look at all those details and lines. Beautiful. We have the mountains in the background. I really like this one. I'll save it for this prompt, the meaning of life. I'm impressed with blue willow. We can actually out paint this image even further and I think that would be interesting. Let's see here. I'm disappointed because as you can see it, it didn't continue with the element here, with the trees or with the grass. It just added some frames, even some text. This out painting wasn't successful. Okay, I've tried another one. I clicked twice on the out painting. It generated the second one again. Let's check this out. Maybe this one is better. Again, as you can see here, we've got some frame on the fore front as well. We've got the frame, the second one and the number three here. It actually added more details, Not too many, but it expanded the image. I don't think this looks natural here. If we go here, this is better. I will choose number three here. And let's expand it. Let's expand number three. Here we have it. 33. BlueWillow Image to Image Generation: This is where we left off from the last video. Now I want to show you how to do image to image generation with blue willow. Here, there's no extra button to do to image generation. You'll still have to start with the slash image here, where you write your prompt, you will add the link to your image. For example, if we just go to Google, let's search for cats images here, for example, let's choose some good image of a cat. This one is pretty here. I can copy image address. Let's make sure that it's working. I'll paste my address here. And it should lead me only to the, not to the website, but to the image. This is something that we can past. Our prompt here, can image address here. Now we can write a prompt. A small it, a small kitten for example. As you can see, it took the information from the image. If we open this image again, you can see a kitten here. Now I pay attention to details. Look at the fur coloring and eyes. If we go back, you can see that I tried to use the same colors. The eyes, you can see like green, bluish tones here. It actually used the information from the link to create these images. Now you ask me, how can I use my own images? This is super simple. Basically, you just need to convert image to a link. There are different ways you can do that. One of the ways you can just upload image here, then here you choose the image that you want to upload. For example, let's use the ballerina and just click Enter. It's going to upload the image to the Discord server. Now if we click on it here, if we click on the right button here, we can have copy image or copy image address. This is what we want. We want the image address. Let's copy image address. Let's try it out. Instead of this kitten. Let's paste our image here. As you can see, we've got address to this image now. Again, put the imagine, put our link to the image. We can put something that we want now, ballerina in a magical forest. Let's add the negative prompt. Because I don't want any extra limbs. No extra extra extra for aspect ratio. Basically, it can be either horizontal, vertical, or square. By default, it will be a square, as we can see with this kitten. It's not affected by the source image dimension. For example, for this one I think the portrait would be the best one. So I'll put the aspect ratio of two to three. I will use version four, version 3.4 They can generate any aspect ratio from the source image. However, versions 1.2 they will be the same format as the source image. Okay, lets zoom in. As you can see here, proportions are not too bad. The background needs some more emphasis. However, overall, it's fine. These shoes are not drawn properly. This prompt is a bit short. Let's make it a little bit longer and add all our stylize, Usually stable diffusion likes those words. I will, I'll just copy this whole thing and add extra information. So again, imagine, now we want to emphasize the magical forest. I'll put the parentheses here. I will also describe colorful trees, leaves, flowers, grass background. I will add styliz, highly detailed. I will also add Unreal engine. And then eight K, and then again, no extra limbs and same aspect ratio and the same version. Okay, here again we're getting this white background. I think the importance of the image is way higher here because we have this white background again. Let's try one more time and we'll put the white background in the negative. Prompt again. Let's try it, no white background. Let's see if this will fix anything here. Here as you can see, we're still getting this white greyish background. The effect of the image is way higher than of the prompt. In that case, Lexica was way better because we were able to generate the magical force background here. We can't Blue willow still needs to work on those features in the future, there will be ways that we can modify the prevalence of the image. Okay, let's do one more image and let's try my portrait. I will upload my image. This one. Let's upload it now. Let's copy the image address again. Imagine I'll put an oil painting of a young woman curly hair. And then I'll add a symmetrical face cubed make up and stylizi detailed realistic eight K. Let's try that. Let's no poorly drawn faced the aspect ratio. Let's have a portrait aspect ratio. I'll put 23. Let's do the version four. Here we have the image. Actually I forgot our link here. Let's do it again. Don't forget the link. Okay, the images we've got here are not too bad. All quite similar. It's just the facial features are a bit different. But given that we only gave Blue Willow one image, like one photo of myself, that's pretty good quality here. Especially the number four though they're not much resemblance with my face. But I think it's just because it's only one image. There's not much information to work with in this task. It did pretty good here. Okay, that's it for blue willow. I think we've covered pretty much everything here. We've started with the blue willow server, We've talked about all those different things in the left column here, and then we've talked about how you can make it work in the direct messages. We've talked about different commands, they can imagine info and subscribed. We've talked about different parameters. Parameters. You can find them in blue willow. They're pretty much quite simple. Negative command, aspect ratio and versions. Maybe there will be more parameters in the future and you'll fight it here. We've tried different art styles here. Try it out, and if you don't have a Discord account, it's worth creating one and trying it out. In the next module, we will cover another platform on Discord. And I think you should be familiar with that because it's mid journey. See you. 34. Midjourney Introduction: Hello everyone. Have you seen a viral photo of Pop Francis in a stylish, white puffy coat? Or maybe Trump being arrested or jive basis cleaning a hotel room. Well, all of these photos have two things common in them. One, they're fake photos and second of all these, all photos were generated by the same AA image generator that I'm going to talk about today. And it's Mid Journey. Okay, mid Journey. Mid Journey was developed by a company called Mid Journey, Inc, which is a San Francisco based independent research lab. It was launched in open beta mode on July 12, 2022, Not quite a long time ago, It only operates on Discord. Now you would need to have a Discord account to be able to use Mid Journey. Already, the company released multiple versions of its algorithm, and the latest version is 5.1 Mid Journey gained a large popularity, and it's already been used for magazine covers, including a famous magazine like The Economist. It's also been used for book illustrations, comics and much more. And now I want to tell you a little story. You probably know this painting, a famous painting of a girl with a pearl earring by Johannes Premier. It's located in a museum in Hague. What happened was the museum, they loaned this painting to a different museum for the time being, they decided to launch a competition to replace with other artworks painting they called this competition. There were about 3,500 submissions, there were only five winners. And imagine what one of the winners was. An image generated by Mid Journey, and it was sent by AI artist, and he submitted it with the title, A Girl with glowing earrings. Out of the 3,500 submissions, an image generated by Mid Journey was chosen. So now you would see this image in the Hague Museum. And as you can see, the image quality is incredible. And that's what makes it my favorite program platform for image generation. And I'm very enthusiastic to tell you all about journey and show you the tips and tricks, how to use it. Let's talk about why I like journey so much. Let's talk about prose. As you've seen already, Journey generates very high quality images that are also realistic. It's hard, or you can say, impossible to distinguish between a real photo and journey generated image. That's how realistic it is. Also, journey is great for beginners. You can use prompts and it will images. It's not going to have cropped images or things like that. If you use short prompt with stable diffusion, it's likely you'll have a head crop. The facial features would be all incorrect. That's why you need longer prompts. You need to add stylizi, highly detailed, all those smaller words, and maybe also add artists to make sure the facial features look good. But with mid journey, you don't need to do any of that. You can just put a girl and mid journey will generate amazing images of a girl. Of course, if you're looking for a specific images, then it's best to kind of elaborate what you're looking for. So in that case, the image that's generated by mid journey will be more aligned. With your own vision. But if you are thinking or if you're looking for some concept, just ideas you want to brainstorm, then you can just put short prompts and that will help in your brainstorming. Okay. It also has many parameters and settings. If you are advanced with mid journey, you can generate and get the results that you're looking for. Some things that you can do with journey is you can upscale images, you can create variations of images, and you can blend images together. You can also do image to image generation. What I think the advantage with mid journey is that you can use many images in your prompt. Also, you can generate images in a private mode if you have a pro plan, which I think is important for some AA artists. Another cool feature about mid journey is that you can use M in your prompt, just modes, and it will make images based on your modes. Okay, now let's talk about some limitations of mid journey or some things that journey may not have. Well, first of all, it requires a Discord account. But hopefully from the Blue Willow module, you've already got your Discord account and you're all set up. But for some people, getting a Discord account may be a challenge. Another thing is that recently the free trial was disabled. And they explained that because a lot of people abused the platform, they tried to find loopholes to generate many free images, and so they closed this free trial. Another thing that I find annoying with mid journey is that it can be quite challenging to generate consistent images or characters. What I mean by this, for example, stable diffusion. If you use a seed, then the images you get are quite consistent. Or you can also train your own models and your models. If you train models with your face or with the character that you want, then you would get consistent images with that character or your personal images. That's impossible with mid journey because you cannot train anything here. I think that's a big limitation with mid journey. Hopefully in the future, they can add this feature so you can train with your images. Of course, you can try to create consistent images and characters with mid journey, but it just takes a lot of time effort. And also you need to know all those tips and tricks, how to do that. We'll talk about that as well. There is another app called, Let me Show You Inside Face. I also want to cover this in our module, because here you can actually put your face to journey generated images. Again, you would need to use another app to be able to get consistent images of yourself, for example. Another limitation is that there is no image editing, so you cannot do in painting or out painting, which I think is a pity because sometimes when you have a nice image, maybe there are certain things like maybe hands that needs improvement. And it will be nice and easy to do within the same program. Just like in painting, however, mid journey doesn't have that. That's for mid journey. In the next videos, we will cover mid journey. Go over how to start using it and all the cool stuff with it. See you soon. 35. Midjourney Overview, Setup, and Basic Commands: In this video, we will start exploring Journey. And I will show you how to get started with Journey, how to set it up as well as I will show you basic commands that you can do with journey. So let's get started. First of all, you will need to sign up or sign in to your Discord account. So hopefully by now you already have a Discord account, so all you need to do is sign in, then go to the official Mid Journey account, Journey.com Here you'll see a button called Join the Beta. This will redirect you to Discord. Here you'll need to click this, accept invite, click this, continue to Discord. Since you're already signed into your Discord account, mid journey server will get added to your Discord account automatically. This is how it will look like. This is their logo here, you will see this is the channel of course. On the left hand panel, we have a lot of different things going on. First it's announcements here, you can check out the announcements. Then we have recent changes, for example, like the changes in prices or changes in the algorithm or maybe a new version that's coming up. It's good to be up to date with. Then status rules for example. Here are some high level guidelines and so on. Terms of service, if you want to read on that a little bit more and so on. Getting Started Guide, but this is what I will show you now. Don't worry about that too much. Okay, here's a lot of information, some support, and so on. But what you're mainly interested is this newcomer rooms. You can click on any of them. For example, new B. Here you can see different images that were generated by others. You can try to generate your own image by going to the window and writing Imagine. But probably because they ended the free trial, you won't be able to do that if you click Imagine. First of all, you probably will need to accept their terms and conditions. That's first, and then they will ask you to subscribe. Let's do that. Let's first Subscribe, and then you can Subscribe Subscribe button here and then click Enter. Okay. Here, Mid Journey will generate a personal link. We can open this page. Yes, it will open a page. Here in this window, you can see your plan if you have any. For example, I have this basic plan. I can see some of the details. For example, how many hours are included, how I've already used up information about the billing and payment. If you probably will see this as the first thing, because you don't have a plan yet, You can choose between three different plans, Basic plan, Standard Plan, and Pro Plan. For Basic Plan, you have limited generations. That's around 200 per month. Again, here for example, the Standard Plan has the generations in hours. The Basic Plan for example, the Standard Plan has 15 hours. The Basic Plans, 200 generations per month is about 3.5 hours. Just to give you the rough idea here in the basic plan, we also have general commercial terms, access to member gallery, optional credit, top ups, and three concurrent fast jobs. Okay, how is that different to standard plan? So here we have way more fast generations. We have unlimited relax generations. The difference between fast and relax generations is that the relax generations, it takes way longer to generate the image. Okay, so here it's unlimited here. Then we have the same general terms, commercial terms, and so on. In the pro plan, we have even more fast generations, 30 hours. Again, we have the unlimited relaxed generations. We have this stealth image generation. You can generate images in a private mode, which I think can be important to some people. Then we also have the 12 concurrent fast jobs. It will generate 12 images at the same time. That's basically what this means, okay? So after you choose your plan, you just click to buy the plan, for example, here I can upgrade my plan and so on. Then you just fill the payment information and pay. Okay, once you have this plan, let's go back to our Discord account. Once you have your plan, you will be able to generate images in this new B group. But what I would recommend you is, again, the same thing that we did with Blue willow is to add the journey bought to your direct messages. How to do that? Again, we just need to click on this mid journey pot. Here we can write a message to this journey bot. For example, High. This will bring me to direct messages with this mid journey bot. Since I've already been using mid journey with my direct messages, I have a lot of images here. Another way you can add mid journey bot to your messages. Direct messages is okay. Let's go back to this journey server. Here you can choose any image. All you need to do is to write. Click on it, you can add Reaction. Click a reaction here. There's a lot of different reactions you can add. But what we're interested in is this envelope mog. How you can find it, you can search for envelope here, you have all those different envelope mog. What you're interested in is this basic envelope. Just click on this. When you add the envelope mog here, add this image directly to your direct messages. As you can see, I went to my direct messages. Here it is. The image that I reacted with this envelope is now in my direct messages. In this same scenario. For example, if you like any images that were generated by others in the same way, you can save them by basically clicking this envelope, MOG. Okay, now let's try to generate something. For example, let's do slash. We can either choose slash, imagine, or we can write it ourselves. Let's write it, Imagine. Now here we need to write our prompt for jury. It can be very simple. I will start with a basic prompt. Like a girl, a girl. And let's click Enter. Now we have absolutely beautiful images of different girls. I think these few are quite dark in terms of the background they have and so on. But if you try writing the same prompt every time, you'll have different, different backgrounds, different facial expressions and so on. So in this case, journey is amazing. As you can see, none of the images have any problems with facial features. For example, or the head is cropped no compared to stable diffusion where if you have a very short prompt, it's likely to give you some problems with eyes, with other or maybe hands also, it will maybe give you the subject off center. So here you can see the girl, like on all of these images is in the center. This is how the portrait should be, but in stable diffusion, as you remember, we were getting a lot of images that were off center were cropped, maybe head was cropped and so on. Here in my journey, it doesn't have flaws with basic prompts which is incredible. Okay, so now that you have this imagined command, I want to show you a few other commands. So here's a list of basic commands. I've already showed you the subscribe command and the imagined command. Now let's check out a few others. Let's do info slash info, for example. Here you will get information about your account. You can see here I have this basic subscription here. The information about how I'm generating the images, I'm using the fast mode. It can be either fast or relaxed in some plans. Relaxed in the standard and the pro plan, the relaxed mode is unlimited. Okay? The visibility mode is public only. The pro plan allows you to do private image generations and then I have this fast time remaining lifetime usage and so on. So it's a basic command to find out about your subscription, what kind of mode you're using, and so on. And maybe when it expires and renews. All right, and this is for the slash info command. Then I want to show you the slash settings, the settings command. This is the one that allows you subsetting. This command actually allows you to change the mode or visibility mode. If you have either the standard plan or the pro plan, instead of this fast mode, you can switch it to the relaxed mode. If you just click on it, you'll be able to switch. If you have this appropriate plan for me, I'm on the basic plan. It gives me this message. Your current membership plan doesn't include relaxed mode, so I would need to upgrade my plan to be able to use this. Similarly, if you have the pro plan, you will be able to switch from public mode, private mode again. You'll be able just to click on it and it will switch again. I cannot do that because I don't have a pro plan in the settings. There are other things that you can change, we will talk about that a bit later. Here you can change the version. Currently I'm using the 5.1 version, which is the latest one. Then there's also different styles. The G is the anime model and so on. The Stylize, we will talk about that as well. Now I want to show you how you can use the help command. If we go back here and write help here, you'll have all the information to get you started with mid journey. It has great resources. For example, the first link is to Mid Journey Docs. If you open it here, and let's click on this quick start guide, here is the information how to set up your account. Similarly, here, you will find all the information about the parameters, settings, and basically everything that you need to know about mid journey. Now I want to go about to subscriptions. Subscription plans, if you are thinking which plan to buy, then you can go to the subscription plans. Here it shows you in more details what are the differences. For example, how much extra is the GPU time. For example, all of these plans, it's $4 per hour and so on. For example here, stealth mode, you can read more about the stealth mode. It's journey is an open by default community and all image generations are visible at Journey.com Including images created in private, discord, servers, direct messages, and on the Journey web app. Right now, the images that I'm generating are all public. They can be accessed by other people. If you don't want your images to be public that you need to buy pro plan and use this private mode. If we go back here here you'll find more useful information. For example, the Mid Journey app. Here you'll see the gallery, the community gallery, and images that were generated by others, for example, let's try that. This is all the images that I've generated. If you go to explore, here are the images generated by community. You can find some inspirations here. Basic commands, we've already talked about that. Imagine info and subscribe the direct messages. How you can add the journey bought to direct messages. We've talked about this envelop emoji, but you can also react with other mom. If you react with this cross, then it will cancel or delete a generation basically. For example, here I have a girl, let's say I want to remove it for some reason. I can go and right click here again. I choose Add Reaction. Okay, let's try Cross here. I don't see the Red Cross Mogi, I'll just try the X. Okay, here we go. This is the X emoji. If I click on this generation got deleted again. You can use these ones if you like. The image, you can react with the star. That's it for this video, I've showed you different commands. The subscribe command allows you to check out plant subscribe and manage your plan. Imagine we'll generate images Info. Allows you to check out your plan information. Settings allows you to configure your settings including to change the model version, Stylized Value. Here we've talked about how to change between the public to private mode and from fast to relaxed mode. Of course, if you have any more questions or need extra resources, you can enter the help to get extra information. 36. Midjourney Text to Image Generation: In this video, we will continue exploring journey. Here I want to start with basic image generation and show you some features, parameters, and also more basic commands. Okay, first of all, again, let's write imagine slash imagine. For example, here I decided to use a very simple prompt and its universe in a bottle. Here we can see four different creative decisions for universe in a bottle. They pretty much look quite similar here. Here as you can see, you have these four images, You have the upscaling, you can upscale any image. This is again similar to the blue willow here. The first image is assigned with the one, this is the second image, third and fourth here. You can upscale them in the bottom row, you can generate versions, You can generate a different version. For example, here, none of the images are quite what I want. What I'll do, I will redo this prompt again. I will click on this pattern. This allows me to change prompt if I want to. I can put a new universe in a perfume bottle. This looks more interesting, I would say number two. The perfume bottle is strange here. And the same with the number four. I'm not sure what this item is, but number 2.3 looks good to me, and I want to upscale the number three. So you can upscale number three, for example. Okay, here we go. The upscaling in this version is actually very fast. And that's because the images that were generated here with the version 5.1 they are already fully rendered. They just need to be separated. This upscaling command basically separates this image from the other ones. However, for other models, upscaling is a little bit different, and for that you need to go to the mid journey dogs. Let's go and explore that because I think this is important because models will change. But just the scale of using mid journey dogs will still be relevant if we go to do journey.com Here's the getting Started here, user guide. Here we have the scalars commands, parameters. We will talk about the commands and parameters later, but for now we're interested in the up scalars. Let's click on this here. As you can see, these are different models for version five. There images that you're getting are already the full size images. Thousand 24 by thousand 24. Version four, for example, the grid images are half of that, only 500, 1,212 The upscaling actually makes the images bigger and it can add some details. Okay, now that you know that we can go back and try something else, let's use, imagine, let's put a portrait of a ring called Tame Farmer. Let's put elderly, okay, hopefully that's correct. A portrait of wrinkled elderly Vietnamese farmer. And let's click Enter. Here we have these four slightly different photorealistic images of Vietnamese farmer and we see this typical Vietnamese rice field hat on all of them. If you don't want specific items in your prompt, then you can use the negative prompt for that. Let me show you. Let's say you don't want the hat, all you need to do is to click on this Remix button. You don't want to see, for example, this hat. All you need to do is to put no and write your negative prompt. For example you don't want to see at. Let's put hat here. Let's write it again. Here you can see that none of the images have a hat in them. And that's because we've added this no hat, which is a negative prompt. If you don't want to see some element in the image, then you should use this negative prompt. Let's say if you liked any of the images, but you want slight adjustment or little change. For example, I think number two looks very interesting. But maybe I would look at other versions, other variations. I can click on V two and that will give me different variations of this image. Let's try that. You can add some more information if you want. For example, smiling a portrait of wrinkled elderly farmer. You can add smiling for example, here as you can see, we've got slight variations. For example, on the first one the farmer wears a hat. And on the fourth one as well, I think the best one is the number two just looks more natural to me. I would say that the number two is very similar to this image, apart from the person is smiling on the other one. Again, because we made the variation of this number two, this clothing also the facial features were kept very similar in the variations that were created. You can see the clothing style, the hair style is pretty much the same. Now I would like to talk about aspect ratio here. By default we're getting this square. But what if you want a portrait or landscape? Let's say for example here I want to. Let's do variations of number two. I don't want a square, I want a portrait. What I can do is I'll put the R for portrait. It's going to be nine, column 16, or you can use some other aspect ratio. If we go back to the, it's going to be in parameters. Here we have this parameter list. Here are all the parameters that our support we've already talked about this negative prompting. No now the aspect ratio. Here you have this brief explanation. If you want to learn more about the parameter, you just need to click on it. Here you'll have a full explanation of this parameter. Again, we have different support with different versions. For example, for version five you have ratio. You can use pretty much anything that you want. For version four, you have a limitation of one by two to two by one. Again, the G five has a new ratio. Here are some examples. For example, this is four to 54 to 774. I also like the nine to 16 or 16 to nine. This is a common video format or wallpaper format. Okay, let's use the nine to 16 because I think that will look nice here. Nine to 16. And let's submit that. As you can see here, we've got distorted result. That's because in this first image it's been a square. And when we tried to use the variation, it relied on this original size here, it just squeezed it in order to not have the problem. You cannot use the variation. You'll have to write the prompt again. Let's try to write a prompt again. Again, we will to imagine, I'll copy this whole prompt In this way we should not have any problems. You can see that here, even though the aspect ratio is the same as in the previous one here, this image is distorted. But this one looks normal. For some reason, not our negative problem didn't work. Maybe we should add more weight on it. And we will talk about how you can add extra weight for your negative prompt in the future video. For now, let's just try a few more generations again. For example, here the number two is quite nice. I'm going to upscale it upscale, number two. For upscaled version, here are a few things that you can use. You can make variations of this image. Again, it should be the same aspect ratio because as we've seen here, any different aspect Ct will result in distortions here. Again, you can make variations or you can check out this image in the Journey app. If I just click here, it will open in mid Journey app. And here is my prompt. Okay, also you can put it in your favorite if you want. If you mark it with favorite. If you go to Journey app here, go to home. Here we have hot new top and favorited. The images that you've marked with favorite will be listed in the favorited and here as you can see, the image that we've liked. Okay, so just to summarize, in this video, we've talked about simple image generation. For now, we've used the simple prompt and we've covered basic parameters. We've covered the negative prompt that you can use the no if you don't want to see certain elements. Although as you can see in the last example that didn't quite work here, we'll talk about that a little bit later. How you can add extra weight to the negative prompt. We've also talked about aspect ratio. That you can change the aspect ratio of the image. For example, you can use the landscape or the portrait mode. Or if you don't add any aspect ratio, then it'll be square by default. We've also talked about different functions, for example, the upscaling functions and how you can make variations and about the remix pattern that you can generate the prompt again and maybe add some more details if you want. 37. Midjourney Image to Image Generation: In this video, we're continuing exploring journey functions. And here I want to show you how you can do image to image generations. So here, similar to the blue willow. First of all, in order to add images to your prompt, you need to get an image address, for example. Okay, let's try something. For example, imagine here you can paste the image link. If I go to let's say Google, and here I found this image, which I think is quite nice. What I'll do is right click here, You can copy image address. Make sure you don't copy link address because the link address is likely to be to the article or somewhere where this image is part of, but you actually want the image address. Let's try it. Make sure to try the image before using it. Okay, let's write this link. As you can see, this is the link to the image. Now we can use it in discord. Again, we have this, imagine now we can baste the image here. Let's write a few words or something that we want to be in the prompt, for example. So this is an image of a wolf. I'll put wolf, wolf. And I want a special effect. I will add phantasml iridescent of wolf. And let's try this here. As you can see, it actually show displayed the image from the link. Now it's generating the images here, we've got the school iridescent effect. All of these images, they resemble the photo that we've provided. All of them have a similar look to our original image. If I go back our original image, you can see this is zoomed in image of a wolf in the images that were generated. We have very similar wolf here as well. That's how you can use images in your prop. Okay, now I will show you how you can upload your own images. For example, let's upload a file. Let's try using the same plena that we've tried with other platforms. Here's the image of the ballerina. Now I just need to click Enter. Here we have it. Now let's write a prompt again. Again, you start with here, you just need to click on it first will expand the image, then right click and copy image address. Let's paste it here. Now we can describe, for example, a beautiful ballerina dancing in a magical forest. Then let's add some details, Trees and Leif again, we've got this image that we've attached in our product. Let's zoom in and see these images. I can see some problems with proportions. For example, here, I'm not sure if the leg is here, and I can see her feet in a wrong place. Here we have some problem with the arm. This one is pretty good. The second one is also not bad. Just the rotation of the arm. Actually, this is not good because look how long this arm is, and this hand is unnaturally rotated from all of them. The best one is in terms of the proportions of ballerina is the number four. And let's upscale it to check it out even further. Upscale, number four. Let's see. Yeah. As you can see, the proportions are overall correct. We have this beautiful ballerina and we actually can see. Forest in the background. As you remember, it was pretty hard to do in Lexica. You had to use different negative prompt and so on. Here, we didn't use any negative prompt, we actually did it for the first time and we got pretty good results. Then we also tried to do it in other platforms where it wasn't successful at all. And here from the first time it was a success. Now I want to show you a little t, let's say you're happy with this image, but you want to add more details or something. You can actually copy this image address and include that in your prompt. Another thing is when you write, imagine your prompt here, you're limited to only one link. You can include more than one link. For example, you can include 34 and as many as you want, basically. Let's say we've copied the image address of this. Let's paste it here. Let's say you have a different vision for the background. You can go to Google, for example. Here I have a wolf. But now you want to search for this magical forest, magical forest background. Look for the images that resonate with you. For example, this one is beautiful, maybe this one. Now we can copy this image address. Let's copy image address of this. Let's add that to our ballerina image. Address the link. Okay, now let's also add our prompt again. A beautiful ballerina dancing in the magical forest. Let's put some colors, so like her pal. Let's try that. Okay, this is something you will see if there are problems with your link. As you can see, I haven't checked the link before I've pasted it. It's always go to check the image address. Before you paste it, let's go back and copy image address. Let's try to paste it. Okay, here we have this image. Okay, we're getting this arrow. And it's possible because of the extension of that image, let's try to find something else. Or alternatively, we can save that image. If we go back here, we can save image, save image, save it on our, in our download file. Now we can upload it. Upload this image and copy image address. Hopefully that works better. As you can see, indeed it was a problem of this image when we saved it and reloaded and use this image address. Now it works. Sometimes you will have to go through this process. As you can see here, I used these two images. It used the image of our ballerina and of this magical forest. And combine it in these images. Here I find that we're not getting those magical forest results. But of course, you can work on it and try to generate more images. And when we go to the advanced parameters, we'll talk about how you can influence the image that you're generating. Now you know that you can use many different images and add it to your prompt to guide AI in. For example, what subject or what style you want to make the images. It also works if you have a specific character, for example. Now I'm not only limited to one image, I can applaud more photos of myself, for example. Let's do that. When I upload a file, here are three photos of myself with totally different background, totally different make up. This one is without make up very different images. Once we've uploaded them, I can write my prompt. Imagine I will now start adding the image addresses. I will zoom and corporate image address. First one, we have these three images, now I will just describe myself and what I want in the background. For example, a girl with curly hair, green eyes, a modern jacket in a park, for example. Let's try that here. In, none of these images captured my facial features. The hair looks quite similar. Sometimes this would work, other times you would just need to add way more images. Also here, as you can see, the images are all very different. If we actually app the very similar images, then it would capture the face a little bit better. But again, it wouldn't be 100% for that. In the next module, we will talk about how you can swap your face with the images that you get here. For now, you can play around and see how close you can get to your own phase. Let's say for example, you can try uploading more images, maybe six or even seven. Yeah, just play around with it. To summarize this video here, I've showed you how you can do image to image generation with mid journey. And basically you would need image address and you can just paste it to your prompt. You can upload more than one image address and it will try to combine different things together. Also, make sure that your prompt describes the images you're getting, something that you want. 38. Midjourney Basic Commands - Blend: In the last video, we've talked about image to image generations. Here, I would like to continue this topic and show you a different feature last time in order, for example, to create this plena in a forest. What we did, we added two images. We added a link to ballerina image, and we also added a link to the magical forest. Now I'd like to talk about a different feature. It's called blend. What it does, it basically blends two or more images together. Here we can write slash and then blend. Here we have it. All we need to do is to upload two images here, or you can drag and drop your images from your desktop directly. I'll upload them. Let's do the same ballerina and the same forest that you know. What's the difference between the regular image to image generation versus the blend function? Let's use this Bellona and the same forest and see how that compares to the regular image to image generation. For now, let's keep it simple. Just upload the two images and click Enter. Here are some images that we've got. Basically, the blending function will just blend two images together, combine certain features or lights, clothing items, and so on. Here we have this nice purplish and bluish background with trees and ballerina in the front. Okay, let's compare that to what we did here. Here we actually wrote a prop. A beautiful ballerina dancing in a magical forest. Trees, leaves, purple, pink, magical atmosphere. In the blend function, you cannot add prompt. Basically, you just rely on the images that you've uploaded. And let me journey, figure out how it combines them together. Here we've got a little different images. What's the difference between this adding the image addresses in the prompt versus blending them? One of them is that you cannot add prompt to your blend function, which is quite important. For example, sometimes you want to combine the image with the prompt. Want to emphasize the prompt, it also have image four reference. In that case, you would use this prompt image generation. If you only one blend of different images, then you should use the blend function. For blend function, there's actually a little bit of information on mid journey dogs. If we go to mid journey here it's a command command list. Then here we have blend. Here is some information. We can upload up to five images here and we can write our dimensions. We can add the dimensions as well. That's basically all the information here. Let's go back here again. Let's try a few more examples for blend function. You can combine images of different styles so you create a new style. Or you can combine a subject and a background, like we did here, for example. There's a lot of things that you can combine together and create really cool images. Let's try something here, for example. Well, I have this image of a beautiful lady that we generated from our writing prompt module. Let's try that. I have this image of a line with interesting effect. We can also use it. Let's try that. Let me show you. You could upload image three. Image four. If I click image three, I can now upload here if you change your mind, you can delete it. You can add the dimensions. For example, here we have portrait square or landscape. Let's keep it square, Let's try that. It's always fun to see how images blended together. Again, to remind you, we have this lion with cool effect. This effect now we have on the girl, which is pretty cool. Here is one of the examples that we can do with blend function. Let's try something else. Let's again use the. I want to use the same photo, but now I want to combine it with an image of a robot. I will applaud this image first and the robot second. Again, I found this on Internet for dimension. Let's use the landscape. Landscape. And let's go again, we're getting these amazing results. Here we have a highly attractive woman here. It tried to add some details maybe from the roboto. The costume looks very futuristic. Just the whole, everything looks very futuristic here. I particularly like the number four. I think the robotic thread in her hair looks amazing. I will actually upscale it. Let's upscale number four. This is just amazing. For example, if you want to make variations of this image, you can click to make variations. And let's see what it will produce. It should be highly similar to this image here. Here we've got some variations. For example, here the probot threats are beside her ear. Here, here we actually, it goes into her head. Not a fun of this one. I think this is random. Again, it's a little bit not structured here overall. The best one in my opinion, is this one. I love how it repeats this hair curve. And just overall composition is astonishing. And this is a very high quality image that you can use for presentations. For example, if you talk about some futuristic stuff, maybe you can put that in one of your presentations. Blend is a great tool to create the images. Let's try one more time. Let's do blend. Again, here I've prepared image of Emma Watson, also a cartoon image. Here I want to show you how you can combine a style with the portrait. Let's use a dimensions, let's use a portrait or square. Let's use a square because I think this is nice and square. It's always surprising how AI would combine the images together. Because for example, here we can see on the first one that it looks more realistic and maybe the eyes were taken from the cartoon. It's always unpredictable. What would you see in the final results? Just worth trying and playing around with. For example, here we didn't quite get cartoonish Watson, but instead we realistic half cartoon style of things. Let's try to generated one more time and see if anything will change or it will create similar style images. Here we've got quite similar style. If we look here, there is not much difference. The facial features will change, but overall, it's very similar style in order to emphasize certain image more. For example, if I want to emphasize the cartoon image here, we cannot write a prompt. We can just add images. In that case, my tip here to add more images of a cartoon. For example, if you want to emphasize cartoon, then add more images of cartoons. For example, here Bland. Let's again add Emma Watson here. Let's add this cartoon image here. I will add one more image. Let's add image number three, for example, this one. Let's try this again. Now here we've got a cartoon style images. If we compare this to the previous images where this is maybe a realistic half cartoonish, Somewhere in between, because we've included one more cartoon image. The cartoon images, We're driving this process now. We are getting the cartoon images. You can experiment with this plant function, you can add more images and just try it around and see how adding more affects the generated images. This is very cool because you don't know what you're going to get. It's always a mystery. For example, for prompts you have the prom to guide your image, so you kind of have an idea where it's going to end. But with blend function, you don't know how it's going to combine those things together. And sometimes it has very creative, unique solutions that can spark your imagination. Play around, experiment. Try different styles. Try combining different totally different images and see what are you're going to get. 39. Midjourney Basic Commands - Describe: In this video, I want to show you a new command. It's describe. Basically it generates prompts based on your image. Let's try that. All you need to do is put describe, you will need to upload your image for which you want to generate the prompts. For example, here I've selected a few famous arts. The first one is a painting by Ali. Let's try. This is a surrealism. Let's see if Journey knows that. Okay, here we have four different prompts here. Surrealistic, grotesque, I think all of the first prompt tells me that it's surrealism. This is very interesting. Number three is the journey that started with the death of a woman for the loss of her son. In the style of visionary surrealism, distinctive noses, surrealist influences. Okay, after we've got these four prompts, you can get the keywords, for example, surrealistic, grotesque, or illusionistic detail, and add it to your prompt. Or you can basically just copy this whole prompt, the one that you choose, and paste it in this window. Let's choose something interesting. Okay, let's try the first one, because I think the first one is the most accurate description. Let's see what journey can generate with this prompt and how close will it be to our original image. Let's copy that. Imagine, and let's paste our prompt. As you can see here, we've got some surrealistic elements here. Overall, it does feel a little bit like deli in terms of the colors and the overall composition. I think here the prompt was to the point, my only concern is that it didn't identify, that this is time. Maybe it should have added time or clock in the prompt. Okay, let's try a different image again, Let's put the describe now I have the Mona Lisa by Leonardo da Vinci. Let's see how Jeri will handle that. Okay, here Jeri spotted on that, this is Mona Lisa by Leonardo da Vinci. In all of the problems, we have the Mona Lisa by Leonardo da Vinci. Then here we have style of oil painting, the style of Leonardo da Vinci. Oil and smooth brush work. Monalisa in Italian painting, in the style of women artists. Here it added more details to it. Other artists as well. Style of Leonardo Da Vinci, Mans Marcel Ucome, classical academic painting. You can use some keywords. For example, if you want to recreate some images that look like from Monalisa, you can look at those prompts and just find the keywords that they're using. For example, the classical academic painting, oil painting and style of Leonardo Da Vinci and so on. Let's actually check out how mid journey would portray Mona Lisa. Let's choose maybe the second one, the longer prompt here. As you can see, it even gives me the aspect ratio. Let's copy that slash, imagine paste the prompt Here we've got images that are in this classical painting style, which is very close to the original style. You can use this prompt. For example, you can replace Mona Lisa with a celebrity name. And use the same prompt to generate a subject with this style. Let's try that. For example, I'll copy this prompt and I'll put imagine. Then I'll write a celebrity name, for example, Zenda. And then I'll copy this whole prompt and paste it here. Okay, I didn't change anything, I just added Zendaya. Let's see if we will see any resemblance with Zenda in the next images. In these images, we can see that the composition, lighting, and the overall classical painting style of the image is very similar to the original Mona Lisa painting. However, because we've added Zenda, it tried to combine the, her facial features with Monalisa. I think here we've got some merge, it doesn't look like either Monalisa or Zendaya. That's what we've got here. You can try to experiment for that as well. Again, let's try another image. Let's again describe as a final image. I chose the art installation by David Tuna. It's the banana with the duck tape. So let's see what journey will come up with. Okay, here we have the first one, a banana with gray tape around it. Very accurate. Okay, and in the style of Jamie, not sure who this is, symbolic object. So I think the first prompt is the simplest and most descriptive one. Okay, let's see. Other ones that we have huge and rose taped banana on a white surface in the style of Mcdonald Punk and so on. Okay, a painted banana with some tape on it in the style. And then a plaster banana with a banana tape to it. In the style of kinetic mixed media, dark gray, conceptual installations. Okay, let's try the fourth. And it has it with the grocery art. Let's very curious what it's going to be generated from the prompt. Again, imagine I'll just base the prompt. Let's check this out. As you can see in most of the images we'll, except I think the second one we've got the realistic image of a banana. It's an art installation, so here we have the wall. It does look like an exhibit. Yeah. Here are some ideas for your next art exhibition. Okay. To sum up, in this video we've talked about described command that allows you to generate prompt from your image. If you are wondering what style your image is, then you may want to use this described command to help you out with the style and also with the keywords. Play around for that and see you in the next module. 40. Midjourney Prompt Writing - Keyword: Hello, hello. In this module, we will continue exploring mid journey. I will show you how to write prompts in mid journey, as well as show you more advanced parameters and commands. Okay, let's get started. The first thing is the prompt writing. We've actually dedicated the whole module to prompt writing with stable diffusion. And we've talked about the organization as well as different parameters like seed and steps. With mid journey, all the parameters will be completely different. It's best to separate stable diffusion with mid journey so that you don't get confused in terms of structure and organization. It can be more or less the same. You can start with medium, then talk about subject action details, background include lots of stylizers, then artists. In terms of the organization, you can keep the same structure even though you can organize prompt any way you want. However, in my opinion, it's nice to have the structure. It's easier for you to write prompts. Let's now talk about keyword weight in staple diffusion. In order to emphasize a certain keyword, you used parentheses. To de emphasize a keyword, you used brackets. Here, it's a little bit different concept. Mid journey uses double column to separate concepts and assign weight to them. For example, we have a hot, do we have the double column dog? What do you think is the difference here? Let's go to mid journey dogs to check if your answer was correct. For hot dog you would have a hot dog, but for double column dog, you would get a dog with some elements of heat, maybe sweat or just the colors are more bright, bright orange. As you can see, the difference is huge because in the first scenario, mid journey treats hot dog as one concept. But the hot double column dog, it's treated as two separate things. This is how you can use double column, two separate concepts. For example, same with ice cream and ice double column cream, because ice cream is a dessert. But double column cream are two different concepts. It's ice and separately a cream. We can try that actually out. Let's use our meat journey to try those two concepts. In the first one, I'll write ice cream. In the second prompt, I will write double column cream. We'll see what are the differences. As you can see when I write ice cream, we are getting different ice creams here. However, on the double column cream, we're getting quite different images. Here I can see maybe two images of the ice cream, but the other ones here we have a woman with dripping frozen cream. And in the first one, it's also a frozen river. We have these miniature balls and dripping cream of some sort. You can see just by adding double column how big of a difference it makes. Okay, If we go back to our presentation here now, we've talked that this double column is used to separate contents, but we haven't talked about the weights. What is it about the weights? If you just use the double column, then you don't assign any weight. It's just going to be a default one, the double column has a weight of one and the dog also has a weight of one. However, you can also increase the weight. For example, here we have cat double column six. The keyword cat has the weight of six. The roof has the weight of one. You can put it in the opposite proof has weight of six, cat has the weight of one. When we talk about weights or emphasis, because cat has way higher weight, that we would see more of the cat in the image. By using higher weight, you indicate that you want this subject predominantly in the image. For example, you should see more of the cat here and a little bit of the roof. In this prompt, you would see a lot of the roof and maybe a small cat. By assigning weights, you can create composition. Okay, let's try that. Let's try these two prompts and see what's the difference. The first prompt is a double column six on a roof Here I'll put one, let's try this one. And the other one is a cat here, we'll put just one on a roof here, we'll put six. When we have a cat double column six or a roof with the weight of one. Here we have a close up of the cat, so we can see that the cat is the main subject of the image. However, if we have a cat with the weight of one or a roof with the weight of six, now the cat is really small. You can see that it just very small portion of this whole composition. And the roof is being the primary object in this image. That's how weights will affect the images. Also, I want to mention that weight, for example here six. The weight of six is applied to everything that's after the separation. Here we have a cat with a double column. One double column is a separator. Everything after the separator has the weight of six. On a roof has the weight of six. The keyword weight is also normalized. For example, double column dog is the same as double column one dog just because if you don't put any weight then it automatically is one. It's also the same as hot double column 20, double column 20 because it doesn't use the specific number, but it uses the ratio between the keywords. Here we have 20.20 the ratio is one. That's why all of these are the same. Keep that in mind when you use the keyword weight. Okay, lastly to de emphasize the keyword, you can use negative number. For example, if you don't want certain things in your image, then you can put double column and then negative in the number. For example, negative two. You don't want to have deformed elements in your prompt, you put deformed double column negative two. That's emphasis on the negative prompt here when we talk about the negative prompt using the parameter no, it's the same as if you used the double column negative 0.5 If you want to de emphasize a specific keyword even more than the no allows, then you can put a higher number and that will de emphasize it even further. Okay, let's strike an example. Let's write a prompt for a portrait, for example. Imagine, let's start with the medium. It's going to be a portrait photography. Then we have our subject. It's a girl with freckles. Then we have the action, she's holding a fox. Then we have a background in a beautiful forest. Now we can add the styliz, such as high quality, award winning photography, aperture, lighting, and so on. Let's put some of them here. Here put high quality, artistic, award winning photography, aperture of 1.8 and natural light, here I put high quality. You don't have to put high quality because mid journey already produces high quality images. But why not? Lighting is important and angles are important. These are the elements that specify how you want your image. Okay, now let's put things that you don't want to see in the image. For example, I don't want black and white. I don't want to have it deformed, and I don't want any watermarks. Now, let's put the weights because right now everything is together. The portrait photography and our subject should have the highest weight. Let's put portrait photography with double column and let's put a weight off to then a girl with freckles. Let's put that maybe a four, double column four, holding a fox in a beautiful forest that also should have quite high weight. Maybe I'll put a three here. Then we have our stylizers and negative prompt. We have to separate the stylizers with the negative prompt Even if you don't want to assign any weight to your stylizersn, need to put the double column. Let's put the double column here, then space. Now our negative prompt is separated. Then for negative prompt, we need to assign the weight. Again, double column. I'll put minus three because I don't want to see these elements. Now we have a prompt orchard photography with the weight of two. A girl with freckles with a weight of four. Holding a fox in a beautiful forest has weight three. Then we have our stylizersn. We have this double column that separates the prompt with the negative prompt. And for the negative prompt, we have black and white deformed water colors with double column and the weight of negative three. Okay, let's try this out. Let's see, we've got some gorgeous images here. As you can see in all of the images we have our girl, I'm not sure if she has freckles here. Maybe we should put more emphasis on the freckles because maybe slight ones but not apparent ones. Then we have a fox here and everything that we put in the prompt. 41. Midjourney Prompt Writing - Option Set: In mid journey. Actually, you can save a part or all of your prompt if you would. It's mainly useful if you generate images with the same style, Such is portrait photography, let's say if we want to generate another subject in the portrait photography medium, Let's copy this whole prompt. Let's paste it here. We would change a girl with freckles. Who? Something else? I don't know. Let's put a boy with freckles, for example. Let's remove the holding folks in a beautiful forest, let's say. Or in a gym, let's put it in a gym. As you can see, you've changed the subject, the action, and the background, but everything else stays the same. All the stylizersre pretty much the same because these stylizers apply to the medium, to the portrait photography also. The negative prompt also works here as well. In order to save time, you can actually save these things in a command. Instead of adding all these words, you can add something as simple as photography. You would get all these styles and negative from within this command. Okay, let me show you how you can set it up. There is a feature called Prefer option set that allows you to set options. This is the command name that you want to give to your specifications. For example, here we can put photography. The value will be all the words that you want to be part of it. Let's try it out. Okay then and here we have prefer option set option will be the name you want some short name and maybe descriptive. These stylizatrait photography you can put like photography for example, photography. Okay. The option is here and then we click on plus one more. We choose value for the value. We choose everything that will be repeated when we use the portrait photography, which are the stylizers, the negative prompt, here we have it. And you can also add parameters, for example, if you want, every time you generate portrait photography, if you want a specific aspect ratio, you can also add the aspect ratio, for example, four to five. After you're satisfied with the value, you click Enter. Now this option was saved. Now you can use the command Photography that will add all of these words automatically to your prompt. Let's say again, we want the portrait photography. Let's copy that. Portrait photography, a boy with freckles in a gym. Now, instead of adding all these Sty lasers, we can add the command photography. Again, you'll need to use then the name that you gave. In my case, I give it photography, so I'll use the photography. Okay, let's see. As you can see, it automatically changed the photography command to the words that we've used. Here we've got the images and as you can see, all the stylizers, negative prompts were added. Similarly, you can save other stylizers and parameters for example. Another example is if you want a character then you can use the stylizers. Digital Art three D rendering real engine. And then for negative prompts you can put deformed or simple. Then you can use the character command to add all of these words to the end of your prompt. For example, here if I put fantasy L and then character, it will add all these words to the end here. In this way, you can create different stylizersfferent negative prompts for specific mediums or specific compositions. For example, when you set up options, you can check them out by using preferred option list. This will show you all the options that you have. If you have a lot of options, this is quite handy if we go back and we put Preferred Option List. Now we need to click Enter. Here, we'll have. Options that we set. For example, here we have only one. Because we've set only one option, we have this photography, here are all the words that are part of it. In a similar way, you can add more options. If you want to delete option, all you need to do is again use the preferred option set here. Instead of setting a new option, you can choose the existing options. For example, you can choose the photography here. You don't choose a value, but you just click Enter. As you can see now it says that our custom option photography was removed. Another thing that I wanted to talk about here is an alternative prom structure. Instead of using the medium subject, action, details, background and stylizers, artists. Here we can change our prom structure a little bit. It's more convenient to save everything including the medium right now. For example here we have only stylizers can also include the portrait photography. We can re arrange our prompt a little bit here. Let's change it up. I'll just copy this whole prompt here. I'll paste it in this window. Let's rearrange it. Let's put our subject first. Move the portrait photography. We have our subject, We have a description in the background that will probably change with every image you make. However, the stylize would be in one place. We have this medium stylers and negative prompt and then parameter. Now let's just remove the subject and the background. Now we can save as the option, let's copy that and let's set it up. Prefer option set, let's call it again. Maybe photography, let's put our value and let's based everything here. Again, we have the medium stylizers, negative prompt, then parameters. And now click Enter to set it up. Okay, now it's set up. We can use it with any image. For example, we can imagine, now I can put any subject here. For example, a mice in a house. I want the photography setting. I will use photography. Then click Enter, because all of the words are ready there. Here we got a photograph of a house, maybe there is a small mice somewhere here as well. Overall, this is how you can save the styles with the stylizers and negative problem together. Anytime you want to, for example, create portrait photography, all you need to do is to write a subject and then use the command that you set up. In my case it was photography and mid journey will automatically add everything else. 42. Midjourney Prompt Writing Resources: Here I've prepared a few resources that can be helpful for prompt writing. For example, this journey styles and keywords. We've discussed this website when we talked about stable diffusion, prompt writing, but I think now it's more relevant. So this is a Github website. Here you can see a lot of keywords. For example, if you're interested in lighting keywords and you just want to explore what kind of lights are there and what would be most useful for my image. You can click on it once it's loaded. Here are different versions. You can select the version that you're using, for example, four V five or Ige for anime. And illustrations here are also different lights, for example, types of lights, lamps and tubes, types of lasers and so on. Find something that's more relevant, for example, types of lights. Here you have different keywords and you can see what image can be generated with this keyword. Right now we're using the version for these images were generated with the version four, for example, air glow or Alp. And glow here, there's a lot of different lights. Okay, if we go back here, there are lots of different styles, colors and palettes, Lots of different stuff that can be used for your prompt writing. Okay, let's go back. Here is a medium you prompt tool, and it's pretty cool, so we just click on it. You can start by typing your main idea, Maybe a subject action and details. Then here are all the different stylized artists that you can use. For example, let's for example, try now. Let's choose a style. The good thing about this website is that it has the example images, for example, charcoal style. Now you have an idea how it can look like. For example, if you're looking for a specific effect, you can browse through all these options. Let's choose futuristic. For example, here you can actually add the weight. The default one is one, but let's say you want a higher weight, so you can move to two, let's say. Then continue. You can repeat that. For all the other Stylis artists, for example, you can choose Gusto Clem. The only limitation here is that these artists are limited. You can see there's not that many. You cannot add your own here. But at the end you actually can add it to your prompt to yourself basically. But you just cannot choose the artist that's not part of this template, okay? Once you choose all these parameters like size, for example, aspect ratio, vertical quality, deol, let's say colors blue. Now you can copy this prompt. Let's paste it to mid journey. Imagine some strange randomly prompt here. Let's try it out. Here it, the version four. If you want to change, you can put version five or you can just delete this parameter altogether. Okay, let's try this out. We've got very interesting results here. I don't see any cats here, but in our prompt, the futuristic has the weight of two. Maybe it interprets as the name of a woman, for example. Okay, now you know how you can use this prompt tool to help you generate prompts. 43. Mijourney Parameters - Image Weight, Quality and Stop: Now let's talk about some parameters that you can use in mid journey. The first one is the image to prompt weight, as I call it in mid journey. It's just called image weight. It's a parameter with the values range 0-2 The default one is one. Basically, lower image weight values less than one and bigger than zero means that the text prompt has more impact. The higher image weight values bigger than one and less than two, that means that image prompt has more impact. If we see this example, this is the image prompt. This is the prompt example. This is the image birthday cake. This is the 0.5 image weight. As we can see here, this looks more like a cake rather than flowers, as we can see here in the image. But as the image weight increases now with the image weight of two, which is the highest value, we're getting something that really similar to our image prompt. This is how you can use image weight to manipulate how much of the image you want to see in final results. Let's try this. Here I have an image of Monalisa by La Da Vinci. I will use that as the image prompt. Let's copy the image address. Here I have my image address and I'll put Zendaya. Now let's try the image weight of 0.5 This is the first prompt. Then in the second one, we'll copy this whole thing, but we'll use image weight of two. We'll see how that compares in the first prompt here, where the image weight is 0.5 We can see quite contemporary images or even photographs of Zenda because here the emphasis is on the text pro. On the second example here the image weight is two. Here we can see that the whole composition, the hairstyle, the clothing, looks very similar to our Mona Lisa image, that what the image weight does. Okay, let's move on to the next parameter, quality. Quality, I would say is somewhat similar to steps, number of steps in stable diffusion. Because the lower the quality, the less details the image has. When you increase the quality, the more developed is the image. However, compared to steps in my journey, you can only have the three values here. It's 0.25 0.5 or one. The default one is one that basically means, for example, if you're doing maybe abstract art and you want less details, then maybe you should decrease the quality, make it 0.25 An important thing to say here is that quality doesn't impact the resolution. You'll have the same resolution as the other image grids as the default image grid, for example. It also affects how long the image will be generated. The lower the quality, the faster the image is generated and less GPU it uses. If you want to read a little bit more on quality, you can go to Parameters here. You can go to Quality and read more about quality Here, let's try something out with Quality. For example, imagine let's put robotic arm. And Quality is, let's use the lowest, the default, 1.25 If you want the default, you don't need to put any quality. Let's just use that one. When the quality is 0.25 the images we're getting, it's pretty detailed, but you can see this little bit. Noise It's not as sharp, but if we look, when we use the default parameter, the quality is one. Here, you can see all the details. If we zoom in, you can see outlines of the wires and all very clear and sharp. Okay, so we've discussed quality. Let's move on to the next parameter. The next parameter is Stop. What it allows to do is to stop the generation part way. For example, halfway. If you use Stop 50, for example, it accepts values 10-100 and default is 100. Stop 100 will give you the default image. Anything less than 100 will give you underdeveloped image. If I can say it that way, we'll have more smooth lines. If you use low stop values, then it can be blurry. Let's try it out in our journey. Let's imagine I want to try something fun. So we can put a Safari hat in a jungle. Now we can use the stop parameter, stop. Let's use ten. Let's now copy this whole prompt. Paste this prompt, and let's use 50.80 Let's see, the first one is the stop with the ten. As you can see here, we've got these very blurry images and doesn't resemble anything. Maybe because you know the prompt, you can imagine this is a dog and a hat here. But as we move up, for example, stop 50. We already have our Chihuahua in a hat. But as you can see, the lines are smooth and the background is blurry when we move to stop 80. Now the background gets more detailed. We've got, we're getting more and more details here. This is what you can do with the stop. If you are trying to create maybe more smoother lines, then you might want to use the stop parameter. 44. Midjourney Models: Now let's talk about models. You can get more information by going to jury dogs for example. You go to versions here, it describes different versions and what things they're good at or maybe not good at. For example, version five here, it says it produces more porter graphic generations than the default 5.1 model. This model produces images that closely match the prompt, but may require longer prompts to achieve your desired aesthetic. Each model has its specificity and you can read about H a little bit more in the docs, and especially if there are new versions that says where you should go and check them out. Okay, the version 5.1 currently has a default and raw style version five. As we read, it produces more photographic generations, but it requires longer prompts. The G five is a fine tuned model for anime and illustration styles. The G five has five different versions. It has the default original, cute, expressive, and scenic. If we go again back to our journey dogs, if we scroll down, this is Ng model five, this is the default image. This is the original cute, expressive, and scenic. As you can see, there's slight differences between these versions. Then you also have older versions like version 43.2 okay, where we can change our model. First of all, we can use that in a parameter. For example, if you write, imagine for example, let's use the same juju in a Safari hat. Here you can version, you can put, for example, five if you are interested in the version five, If you're interested in older versions, for example four, you can put version four and so on. Another way you can do that, if you go to Settings Settings, let's click Enter. Okay, here we have different versions. Version 12345, currently I'm using the 5.2 I'm quite surprised because they've updated this version just today. The 5.1 version is now odd. It's important to go to the journeys and read about different models because they update the versions very quickly. Make sure you can follow up with all the different models. Here, I'm curious, let's try version 5.2 Let's compare that to 5.15 and use different raw mode and so on. I will use a prompt. Imagine character mixed race girl in stylish clothing. I will first try the latest version, 5.2 and then we'll try others as well. Right now I'm using the 5.2 When you click on the settings here we can see the suffix, the parameter 5.2 Here we go. Let's try this out. Okay, now I will use the same prompt with a different model. Again, imagine now I don't want the 5.2 I want the 5.1 change the 5.1 Let's enter the 5.1 also has the role mode if you ever get confused how to write it up. So for example, version 5.2 and row mode, here you can see you'll use the modal number 5.2 and then you'll add the style raw. Let's add the raw style. Imagine we have 5.2 Let's add the raw style raw. Let's now try the version five. Let's try the G version five. Again, we'll put the imagine or prompt again. You can go back here. As you can see now it's G four. If we want the G five, it's going to be G five. Let's put that in. Okay, let's check out some models. This is the G four. As you can see, it's more simplistic compared to the G five, but again, it's very different styles. Then we have our version five. Well, let's start with the version four, This is the regular version four, this is version five. We're getting these characters in full body size 5.2 style is in a way similar to the modal five because here we are having full size characters. Then this is model 5.15 0.2 As you can see, the 5.2 has really nice lighting and background compared to 5.1 where it's more simplistic or even white background. Here you can see that with all these models we've got quite different images. It's worth reading some information about the models and trying different things or different styles. 45. Midjourney Parameters - Stylize: Okay, let's go to Stylized Values. The stylized values are also part of the settings. If we again go to Settings here you can see the versions. Let's switch back to the latest 5.2 version. Here you can see different stylized values. You can see a stylize, stylize medium, stylize high, and stylize very high. What are the stylized values? Let's go back here. The low stylization means that images closely match the prompt but are less artistic. High stylization means the images are more artistic, but they can be less connected to the prompt. This is somewhat reminds me of prompt guidance for stable diffusion, But with the prompt guidance, it's opposite. The lower the value, the less it matches to the prompt. However, here, the lower the stylization, the more closely it matches to the prompt, and less artistic, the high it is, then less connected it is to the prompt. Different journey versions have different stylized ranges. How can you check which version supports what range? Well, you go to journey dogs here. Again, if we go to parameters parameter list here. If we look at the stylize here, you've got some ranges right now, version five, version four and G five have the same stylize range, 0-1 thousand. The default number is 100. Let's look at some images. For example, here we have a prompt colorful risograph off. If you don't know what's risograph, it's basically a printing technique for the style zero. We're getting this basic Gu form that does look like a risograph. Here we're getting this basic background, yellow, white, and so on. As we increase the style parameter, now we're getting more details. This is 100 is the default parameter. As we move even further, 250 here, the images start to look less like a Risograph but more as a realistic drawing. For example, here as we increase the styles to 1,000 here we have so many things going on. Very rich backgrounds, very rich details, and so on. As you can see for style zero, we're getting these basic images that are aligned with the prompt. However, as you increase the styles parameter to high values to maybe 750 or 1,000 you start to get more detailed and reach images that may not align with your prompt. Sometimes you want to generate more basic images that correspond to your prompt, and other times you're looking for more creative and artistic images, then you can use a higher stylized value in the settings. Here we have stylized low, stylized medium, Stylize high, and styliz very high. What do these keywords correspond to here? The stylize low is the same as the style 50, Stylus medium is the same as the default Stylus 100 Stylize high is the same as the Stylus 250, Stylus very high is the same as Styles 750. You can set up stylized parameter here. All of your images will be generated with the same stylized parameter. For example, if you want more artistic images, you can use the stylus very high, or you can use the default and write your dial parameter in your prompt. For example, imagine I will use bike infographic illustration, for example, I will put the stylize you need to use style, let's use zero. You can also use, instead of stylize, you can also use the. Then let's put, for example, default 100. Let's check this out with the style zero, we're getting this basic background this big, doesn't even have any keywords. Yeah, as you can see, very basic composition, just the two D illustration. As we move to 100 here, we're getting a little bit more creative. As you can see, images have texts. The first one is interesting, it has the forest background really cool. And then we move to 400, and now you can see that the bicycle is now more three, you can see some cool things inside of it, I'm not sure what they are, but it's definitely more than just a regular bicycle. This is the stylized 400, and when we move to 1,000 now we're getting really crazy things. A bicycle with lots of electronics, wires, a lot of things that were added to bicycle. And on the fourth one we can see some information about house trees and so on. A lot of things. That one is 1,000 and we've got another image with 1,000 Again, too much things going on here. Now you know how you can use the stylus function to change how much of creativity you want to add to your image. Or if you just want to generate basic images that align with your prompt. 46. Midjourney Parameters - Chaos, Tile, Seed and Remix: Let's move on to chaos. Chaos is an interesting parameter to call chaos. You just need to put the chaos or the chaos affects how varied are the initial image grids? It ranges 0-100 and default is zero. This is the default image. As you can see, the images within the grid are quite similar. We have this pink owl with green eyes and green background. When we move to Chaos 80, now we're getting different images in different styles, different medium. Here we have carved watermelon. Here's maybe a plastic owl. This is a tattoo or cool illustration. As you can see, each image is very different. That's what chaos variability. Low chaos produces images within the grids similar to each other. High chaos produces images that are varied and have unexpected composition or artistic mediums. Let's try this out. Let's try a fun prompt, A cartoon lizard in a raincoat walking in a forest. To add chaos parameter, begin using chaos or simply C. Then we can use value 0-100 Let's try with zero, that's the default one. Let's generate a few more. Let's do 50.100 with the default chaos of zero. This is a default image. We're getting quite similar images between each other. Of course the style is a little bit different. For example, here we are getting more illustration here, the background and lighting, it's more three D. But overall, the image and composition is overall similar. When we go to Chaos 50, now we're starting to see that all the images within the greed are quite different. Here we have a cartoon here. Also cartoon but completely different composition. Here we have a moon, a lizard is in the puff jacket. Here. Again, completely different composition. And this is some realistic, a bit scary thing. Okay, let's move on to Chaos 100. Here, I don't quite see a lizard on the second image. I don't see a lizard on the third one. I see a person on the fourth one. This is not a lizard, this is some random cartoon character. The first one, well maybe this character looks like a lizard a little bit with a butterfly ears, I don't know. But as you can see the images very a lot, that's the whole point of chaos. It increases the variability. You can use chaos to look for different composition. If the first images that you generate are not something that you're looking for, you can increase the chaos a little bit, but maybe not too much. Otherwise, you couldn't get like very weird things. Okay, let's move on to tiles. That's a fun one. If we add a tile to our prompt, for example, we can use a simple one, like Music Now we can put the tile, what it does, it will create a similes pattern that you can use for clothing Merge and so on. Here we've got four results. You can choose the style that you liked, for example, the first one is quite nice. Let's upscale the first image. Okay, now let's save the image. We can actually check the seamless pattern. I found the seamless checker here. You can upload your file. As you can see, it works. There are no problems with it at all. It nicely adds to each other. This is how you can use the tile parameter, okay. Let's move on to seed. Here would be quite similar to the stable diffusion seed. However, in mid journey seed numbers are not static and should not be relied upon between sessions. They're not quite reliable before people use them to create consistent images, but right now there are other ways you can do them. Let me just talk about a little bit here. Using the same seed number in prompt will produce similar images. That's the same as with stable diffusion. Let's try something. For example, imagine a fox in, in a forest. Let's generate this here. In order to get the seed number of these images, you would need to go to reactions and add an envelope reaction to the image. Once you add the envelope reaction, you would get the job ID and the seed number. Now you can copy the seed number again to get the same prompt, I will use the same prompt. In a forest, I will add the seed. It's again, it's seed. And the seed number here, it should produce the same result here. As you can see, by running the prompt with the same seed, we got the same results. But the problem is, is not quite reliable in my journey because it can be different between sessions. There is another way you can create consistent characters or modify your pro, and that's the button that we've talked a little bit about. The remix button is a new feature that allows to change subject lighting, add remove elements, and adjust settings while keeping the overall similarity to the starting image intact. For example, let's try a character. Let's, for example, imagine a portrait of a girl with freckles. Let's generate tests here. For example, let's say you're interested in number three. I want to add a different reaction. For example, smiling or a different background. I can add that in the remix. For example, a girl freckles, let's put smiling. I also want to change a background, green background here. As you can see, I think the first image is very similar to the bird girl here. The remix is in a way like a seed in stable diffusion that allows us to modify our prompt a little bit and add different emotions like smiling or frowning. And also change the lighting background and other parameters. I want to show you another really cool thing about remix. Let's use a different prompt. Imagine a dozen eggs in carton illustration. Now we've got some X here. Well, I think all of them are quite similar. Let's use number two for example. You can use the variation or remix. It doesn't quite matter, but if you like a particular style of the image or composition better then use the variation. In this case, I like the number two. I use the variation two. Here I can remix the prompt. I will change the subject. Instead of x, I'll put hamsters, a dozen of hamsters in carton illustration. Everything was kept the same except of the subject. Here we can see that we've got the same composition as we have with the x. Here we have this carton with x. Now the ****** that were for x are taken by the small he hamsters. As you can see, the composition is the same. But now we change the subject to hamsters, and instead of x, it put the hamsters to these places. This is something that I think is really fun because it allows you to use the composition from one image and use it completely different things. Okay, Maybe, let's try a different image. For example, the fourth one here, I'll change eggs to happy owls. A dozen of happy owls. Here we've got some owls. I think number two is the best one. In terms of details, I like how journey tried to make all the eggs here into owls. As you remember, those three eggs were flying not in the carton. These ones mid journey made it into owls. Happy owls. This is something very fun that you can do with the remix function. I think what it also really good is if you have a certain image with the composition that you like, you can use it with a different subject, but keep the same composition. Yeah, the remix takes the general composition of the starting image and incorporates it into the new generation. You can also add or remove parameters such as no style or stop and much more. Also if you remember from the first module on mid journey that when we tried to do remix and change the aspect ratio, it stretched the image basically. If you are changing the aspect ratio, that's what it's going to do. 47. Midjourney Emojis: I decided to finish this presentation with modes. Basically, with mid journey, you can only use modes. For example, you can use this microscope to symbolize the microscope photography. Let's try to use modes with mid journey. I can again use the image and I will use the microscope emoji and the strawberry mode. Here we've got the straw brace, but we didn't quite get this micro photography. Let's try a different prompt. Here I have the microscope pug and a mushroom here. I want to try to use the version four. What I found is that with the version four, if you add the microscope, then it would give a micro photography better than other versions. For example, we can see here maybe inside of the mushroom with some fruits. I'm not sure, but it definitely looks cool. Okay, let's try one more. For example, imagine I'm again going to use the version four, and this time I want to use very different modes. I'll use the rocket and a person in the yoga pause, a man in lotus pose. Let's try that. I'm curious what it's going to generate. Okay, let's add the version four here. We've got an astronaut with really beautiful background. However, I don't see anything to do with the, the lotus position or yoga and you were here. But it's something to do with the rocket and space, that's for sure. This is how you can play around with Mo. You can add emoji and prompt text. Prompt. You can add both things together and see what it's going to generate. It's a fun way to play around with mid journey. 48. Midjourney Image Generation Example- Portrait: In this video, we will try our prompts with stable diffusion, the same problems that we've used with other platforms. Let's get started. Okay, first is the photo realistic portrait. Let's use the imagine. The first one is professional portrait photograph of a young British woman here. As you can see, we don't have any weights. I will add some weights here. Professional portrait photography, that will have a weight of two of a young British woman in a jacket. Let's put young British woman in a jacket with wavy blond hair. Let's put that if four. Then we have beautiful symmetrical phase, cute, natural make up that. Let's put the weight of three. Then blurry, rainy city, street background, that's also important. Let's put the weight of three as well here. Then we have our stylizers, we can keep that the same for the aspect ratio. Let's make it four by five also because I want the images a little bit different from each other. Within the grid, I will add the chaos again. Chaos, maybe not too high, Let's put ten. There's a little bit variation between the images. Let's also use the stylizer will be a little bit more artistic. Let's put the stylize, let's use High Stylize 250. In these images, I actually don't see any street background. And that's because I put the space here between the weight city street background was treated with the weight of one. Because here we have this space. Let's run it again and not have that mistake here, The blurry, rainy city street background. Let's remove the space here while it's generating the image, let's try the other portrait. Prompt again we have, imagine, let's portrait photograph of a young Indian woman with long hair, beautiful symmetrical face. So basically very similar. The background is different, colorful street market background. Let's also assign some weight here. Again, portrait photograph I would the weight of Indian woman. Let's use that as cute, natural make up. Let's put that with the weight of three. And then we have colorful street market background. This is very important. I'll also use the same weight as with the woman. Let's put three here. Let's also add the Chaos. Chaos. Let's use the same ten and stylize, let's use maybe 200. Also I want to use a different model. I want to use the version five. I'll put the five because it's more photo realistic. Let's use that one. Okay, we have these images of a young British woman here. As you can see, we've got this nice street background. The space here is very important. Make sure you don't have the space between column and the number when you're citing the weight. As you can see, now we are getting this background here. Overall, the images are quite nice. I like number two here. If you want to upscale a specific image, you can upscale it, but because it just separates the images. If you want to, let's say upscale all the images, the easiest thing to do is to put an envelope Emoji here, add reaction and find your envelope Mogi. When you put the envelope Emoji, you'll have a new message. Now you'll have the full size images separated from one another. We've also got the images for the young Indian woman, here we. Get a girl here. I think the more appropriate images were third and the fourth one, but the background is not colorful enough for me. Let's try to change this a little bit. I'll copy this prompt again. Let's use this prompt now. I'll put the weight on a young Indian woman with long hair. I'll put the weight here with four colorful street market background. That four, I also want to emphasize colorful. I'll put the color, I will put that also. Four, highly detailed. Okay, let's try with the version five. The version five has less of this journey aesthetic. Let's use that one. Let's use the newest version here, 5.2 for the version five here. Well, some colors, definitely with the image number two and number three. Number four looks a little bit more plain. With the version 5.2 This is what I'm thinking about is having these very colorful clothing items. Maybe like here, but we've got wrong ethnicity here. Let's try again. Let's put five with long hair, that's not too important. Long hair, beautiful symmetrical face, colorful street market background. With weight of four, Colorful, highly detailed, and so on. I'll put the chaos lower. Let's put chaos, for example, two style. I'll remove the stylized parameter at all. It is more aligned with my prompt here. Here we're getting colorful images, but I was looking for Indian as the person from India. Now we're getting Indian as the Native American feathers and stuff like that. In order to direct me journey to the right ethnicity, let's add some Indian traditional clothing like say here. I'll add the clothing attribute wearing. Say here I also specify South India and the woodland. Also, I'll add lower stylize. The default stylize is 100. Here, I'll put stylize 80. It aligns with my prompt a little bit better. The results here are way better. This is something that I was looking for. We have this market in the background and overall, just very colorful palette. We have an Indian woman here in Sari, very beautiful. I would also probably emphasize the background a little bit more because in these two we have quite plain background. But the other ones, I think the number two and number three are pretty good. 49. Midjourney Image Generation Example - Logo: Now let's try some logos. The first one is line logo of cup Kick with a tear and top clean line, simple shape, minimalist vector. As you can see we're getting these very simple images that what I'm looking for in the logo. Very little detail, quite plain background. You can see we've got different colors here. If you want white background, then you can put white background. Let's use it again. But now let's increase the chaos. For example, if you're looking for some logo ideas, I think it's important to add a little bit more chaos. You get different results. Let's put chaos. I think we can put 30. Yeah, let's put white background. Now we're getting these a little bit different results. You can choose the one that you like the most. Also with Journey is nice that you can actually add the text here as well. For example, if you're creating a logo for a bakery, for example, you can put a line logo for a bakery and then portraying cap with a tear in top. If you want a name you can put also at then we want simple shape vector, white background and all use the same caves. Now we're getting not only the image of the cupcake, but also the name of bakery, for example. As you can see here, the words are quite legible, but they don't mean anything. But I think it's a good guide for fonts. For example, this image, a phone like this one, would be very suitable. And here we have this cursive font. This gives you ideas and I think this is a great resource to try it out. For example, I really like this number four. I think this simple and very cute. Let's use a different logo. Let's imagine tree inside a water droplet, slick and minimalistic logo. Ecto graphics. One color white background. Let's remove the color. Let's put two color, two color palette, temper style, eco friendly business details. I would actually put a higher weight here just because we have lots of words. I'll put the weight of two, weight of two, no ****** and space here three inside of water droplet double column two. Then we have logo vector graphics and so on. Okay, let's try this out. Actually let's use sleek and menus logo, let's also put that width two. Now for our parameters we can put chaos. We have some variation. Let's put 20. We don't want too much stylize. We can make stylize a little bit smaller than the default stylize. The default is 100, let's make it 80. Let's try this out. Let's see, this is not bad. However, I think this is too much details for a logo. I will actually change the prop a little bit again. I'll put, imagine here inside a water droplet. I'll, I'll just use the minimalistic logo. Minimalistic and line logo, then vector then that's with the weight of two. We have two color palette, white background taper and so on. Let's make the style even lower. Yeah, let's put 50 here. I think the number three is the best one in terms of the Vector logo here. However, we are not getting this nice. Let's try again. Inside a Water D. Let's use just the line logo because sometimes too many words is not that good as well. Then we have two color palette. Let's try one more time. I'll use the, imagine I will use the tree inside of water droplet. I'll keep that same. I'll change the quality to 0.25 I'll make the chaos a little bit higher so there are more variation. Let's try this out here. The images are a little bit more on what I'm looking for. However, the best way to let me join you know what you want is by having a reference image. Let me find a reference image on Internet. Here I put the loba try in water droplet and here I've got different company logos. I can use something for a reference. For example, this one is quite nice. I'll copy the image address here. And I'll put to my prompt. So I'll use the same probed, but now I will add the image address and we'll paste my probed as a side node. You need to be very careful using other images as you reference just for copyright or legal issues. Let's see these images. I think the second one is nice. We've got this quite simplistic tree, even though we didn't get this water droplet shape. But we can create variations and see if number two, I want to also add what a draw put shape, maybe that will help. Here we've got some variations and as you can see, I think on the fore front we're getting this, what a draw put shape. And the tree, of course, you would need to change it a little bit, but overall this image is not too bad. 50. Midjourney Image Generation Example - 3D Render, Anime, Characters, Landscape and Concept Art: Now let's move on to magical realism and create some three D renders. Again, let's imagine our favorite three D render of the raccoon reading a book. Yeah, let's keep that the same in terms of the aspheric ratio. Let's use a square for chaos. Let's be chaos a little bit higher chaos, Let's use ten. Here we've got really cool images of the raccoon. In terms of proportions, everything looks really good. Here we have this dim light cozy background, cozy setting, and we actually can see a lot of details. If for example, upscale number four here, to upscale the number four here, we can see a lot of objects like shelves, candle box, then we have a lot of other things in the background. It does feel like home. Okay, now let's move on to illustrations and characters. Here I have portrait of skinny anime boy with glasses listening to music in the street of rural Japanese city. Here, because this is anime, the keg model works better with anime. I'll use the keg, I'll put a Eg five if we go quickly go to the Journey Dogs. Here you can see all the different versions. There's the default, the five, but also you have different styles. It can be original cute, expressive sat. Let's try first with the default and also try with different styles. Here I have the keg five. Let's use the same one with the styles G five here I can put cute. Let's do another one with scenic again five and then style set. Here we've got different styles. The first one was the default. We have this Anime Boy, as you can see, change when we add the style style cute, now it's a different style. Then when we add the scenic, then it's also a different style. Depending on what style you want to create Anime, you can choose the corresponding style. Okay, now let's do some characters. For example, a girl riding a bike by our, I have here a girl riding a bike by artists. I wanted to be a character sheet, I'll put the sheet. And then a girl riding a bike. Let's try this out here. You can see that we've got this white background. When you put the character sheet or the character concept art, that will usually give you white background and more of the character. If this is something that you want, then you should use these keywords. Okay, let's try something different. Again, I'll use the, imagine I will, I'll use the character sheet, then I will put a mixed race girl in stylish clothing. I would also add keywords that would be applicable to characters. For example, three D and I'll put concept character art with front and back view. Also, we can add Unreal Engine, that's good for characters overall Unreal Engine. Okay, let's try this out here. We've got characters from different angles as you can see. So we've got the clothing, and now it really looks like a character from a game. If you're developing a character for a game or for a book, you can use Journey for inspiration for example. Or if you're developing character for a book, for example, for book illustration, then you can use these remix patterns to place the same character in different positions with different emotions and so on. So let's move on to landscape. We have here digital art of magnificent medieval castle between the hills and fields, large panoramic background with dense nature and mountains. Grand fortress, Epoced fantasy. Let's try what it will generate and then maybe we can add more parameters for Asper ratio. I want it to be a landscape. Let's put three by two here. We've got some epic images with the panoramic view of the castle. I think for landscapes, if you want to make landscape a little bit more dreamy, then you can use the stop parameter. Let me copy the prompt again. If you want more soft lines or more soft, for example, sky or trees and so on, then we can use the stop parameter. Let's add it here. We have the aspect ratio, let's put it stop. If you remember, stop is 0-100100 is the default and fully developed image. If you stop the generation process halfway or part way, then it's going to have more smoother lines and bluer things. Maybe not make it too low, maybe somewhere around 75. Let's check this out here, we're getting more dreamy atmosphere. And I like this more because with fantasy it works really well compared to the previous image where the lines and outlines are very sharp. Here we're having the more smoother shapes and especially the sky or the background looks more enchanting and light. Okay, let's move on to our last prompt. And it's the meaning of life. The meaning of life. Breathtaking art, stunning high resolution, highly detailed, inspirational, and eight K. We don't need that with mid journey. Yeah, let's try that out without adding any parameters first. Okay, here we've got some amazing artworks. Look at all these details. In three of them, we have a person and just lots of things going on. I want to upscale the images and look at all these details. Let's do that. Let's upscale maybe 1.2 Let's take a look. Here we have a back view of a person. We have all those different elements. We have contemporary elements like cars, sculptures, a lot of things. I don't know if it's a theater or some cool architectures combined together. Very fine concept. If we look at the other image, that's also very interesting. This is in more surrealism style. Again, we're getting all the details. Also I want to add that with the new update that happened while I was recording the scores, We've got new features that allow us to zoom out, basically do out painting, let's try that. Here the image was extended, the boundaries of the image were extended. It's a bit hard to see all the elements in order to see them better. Let's scale. Let's upscale the first one. As you can see, now we have even more details and things going on. Yeah, you can use the zoom out hinting feature. One thing I want to try with this concept art here, we're using the default stylize. I'm wondering what's going to be generated if we increase the stylize parameter to 1,000 This will generate the most creative and artistic images. Let's try that. We'll put the stylize the highest, 11000. Let's see, Again, we're getting so many details. Yeah, this is, the third image, reminds me of the artworks by Dutch painter Bosch. So I'll show you what I mean. This is one of the works by Boss. As you can see, there are a lot of elements here and we're getting also a lot of elements in our images here. The second one, I think this is deep. The fourth one is a fun, dreamy image. Okay, so we've tried all our prompts with mid journey, and for every prompt we wrote, we got beautiful, amazing images. And we've also tried different parameters, such as stylized parameters Aspherio. We've talked about the stop parameter and many more. So now it's time for you to try it out, play around and bring your vision to light with my journey. In the next bonus video, I will show you how you can put your face on any image that you create with my journey. 51. Midjourney Bonus Video - Faceswap with InisghtFace: In this bonus video, I've decided to show you how you can put your face into one of the mid journey generated images. For that, you would need to create a server. So pay attention because it's a little bit advanced, but it is fun. So let's do it. Okay. When you sign in to Discord account on the left panel, you'll see this plus. Click on this plus here, you can create a server. You can create your own, or you can choose from a template. I usually like to choose from the template because the set up is much faster. Once you choose one of the templates, then let's choose for me and my friends here. You can put some image, for example, something like this. Or you don't have to put any image. Now we can name our server, for example, my face swap. Let's create it here. You've got some channels for information, then you have the text channels. You usually want to use the general one. And you can create more channels if you want to. What we'll need to do, we'll need to invite mid journey board as well as the inside face bot. This is the bot that allows you to put your face to mid journey generated images. Let's start by inviting mid journey bot. We've created our server. Now let's go to mid journey here. Right now we're in new B group. You can go to the Mid Journey server and you'll need to find the Mid Journey bought here. And you can click on this ad to server. Or you can go back to your direct messages here in the mid journey board. Also, you can click on this journey board and then click Add to Server. Once you click Add to Server, you can select which server you want to add. We've named our server my face swap. Let's add the Mid Journey bought here, then let's continue. Let's give authorization to Journey bought for all the following. Let's authorize it. Let's confirm that I'm a human. Now we can go to my face swap and that will bring us to our server. Okay, if we go to general here, you can see that journey bought was added to our server. How do we know that? If we go to now we're getting all the commands that we can use with mid journey. For example, we can imagine here such as a grass hopper for example. You can see that we feed our server. We can use mid journey board the same as we used with direct messages. Here we're using the general text channel. However, you can also add more channels and create private channels if you want. Okay, now we have this mid journey board here. We will also need to add the inside phase board if we go here. This is just the link to inside phase I. And here they have all the different projects and information. However, what you really need is this inside phase discord. But if you click on the link here, you can add the bot to your server. Here we've already got our server selected. Make sure this is the correct server. My face swap. Then let's continue. Give authorization, send messages, attached files, and so on. Authorize, confirm. Now inside phase has been authorized and added to our server. Then let's click to go to our server. Now you can see that the inside phase swap was added. Now we have two bots, the journey board and the inside phase swap. In order to use inside phase swap, you'll need to upload your own image. Let's do that. For that you'll use the command here on the left panel. You can choose what bot you want to use. For now we want to use inside phase swap. Here you can see what commands a part of the spot here for example, it has the safe ID. Said ID swap ID and so on. What we're interested here is the saved here. It takes the ID name and the image here. You can upload image of yourself or anyone that you want. Let's use my image here for example, this one. The better the quality of your image, the better the results will be. Here we've uploaded our image here. You need to write the name, the ID name. You can put whatever you want. For example, one. Let's click Enter. Now our ID name is set up. What we want to do now is to generate some image with which we can swap our face. For example, I will use the Mid Journey. This is the Mid Journey command. Imagine here I can put a girl superhero with blond curly hair. Here we've got some superhero curls. Let's choose one image. For example, number two, upscale it. Now we've got this image. If we write, click here, now we select Apps and swapper, choose the swapper. That's it. Here you go. As you can see, the inside face used, put my face on top of the superhero girl face. As you can see, the proportions are quite all right. It captured the features. However, what I found out is that you have a straight looking face, the feature works a little bit better. And also, if you find the image that looks more like, in order to make an image that looks more like you, you can actually use your images when you're creating mid journey image. For example, let's again use this one. I will copy the image address here. Now I will use mid journey. I'll use the image, I'll paste the address of this image. I will describe myself. A girl with dark blond curly hair, green eyes, business portrait photo linked in avatar photo professional. Let's try that here on all the images, the model is looking straight. That's because we've used the reference image. My photo where I'm looking straight at the camera. Here, we've got nice and straight looking photos. That's exactly what we need. We can swap faces with one of them. For example, let's use number four. Now we need to double click, choose Apps, and choose Swapper. Here we go. We've got an image, a professional image, that resembles my photo. That's how you can use your photo on any image that you generate with mid journey. I think this is a fun tool to play with. Try it out on this node. We finish our mid journey module. In the next modules, we will talk about AI photo editing. See you then. 52. New Update: Midjourney New Features: Hello everyone. This is a mid journey update. We will see what's new in mid journey in November 2023, Okay? So let's see what features have changed, improved, or are just new. Okay, now we have a tune command that allows to choose a particular style of images to generate. In, in a sense it's fine tuning and enables us to generate images in a specific style, which is pretty cool because they didn't have any fine tune functions before. Then we have an up scaler, so now we can upscale Journey by two x or four x. Then we have editing, which is also very exciting. Now we can edit specific elements in the picture. I will show you how. Finally we've got this weird parameter that makes images weird and you can try and experiment with it. Let's get started with the tune command and see what it does. Okay, let's go to mid journey Do here. Just put in style tuner here what it says here. It personalize the appearance of your mid journey images using the style tutor. Use the tune command to generate a range of sample images showing different visual styles. Based on your prompt, choose your favorite image and you'll receive a unique code you can use to customize the look of future jobs. Okay, you just put tune and then your prompt if another user has previously generated a style tuner. With your prompt, you will receive a link to that tuner. Click the link to access it. We'll try just that. Let's go to Mid Journey. Here I am in Mid Journey board. I'll just put tune here. We can put either a very simple prompt, something that someone already has done before. Let's do that. For example, photography. Hopefully someone else did the tuner. As we can see here, we have prompt photography. We can see that someone else has created a style tuner using this prompt. Here you can see different number of images, 32, 6,428, It's best to use the biggest number of images because that gives you a wider range to choose from. But we'll start with the 32, just for you to understand how it works, let's just click this, okay. From here you can choose the style of images that you like. For example, I want to generate my images in this minimalistic style. I will just choose this, and it gives me this code. If I use this code, for example, I imagine, and then I'll just put a boy with a balloon. And now I will paste the style. This is the code. It should generate the image in this style. Let's try it out. The images are in the same style as we chose here. This code corresponds to the specific style. Let's say if you chose not this one, we can put that back to the center and choose a different style. As you can see, the code has changed. Now if we use this code, it's going to generate in this new style, but you can also choose multiple styles and it's going to combine them. The more you choose, the more generic it will become. If we go back to the numbers here, the more images it generates in the style tuner, there images you can choose from. Let's say you want to create a specific style that you're looking for. You generate let's say 32 images and the style you're looking for is just not there. Then your best bet is going to 128 images. That's going to give you a wider range of styles to choose from. But say you want something more specific, then you're going to go and tune here. You're going to put a specific prompt, For example, a minimalistic logo. Of angry cat, something like that. And let's press Enter. Now you can see that no one else has the same prompt, hence there are no links for that reason. Here you will need to choose how many images you want to create. Let's choose the simplest 116 style directions you can choose or Default. Now let's use Default and click Submit to start generating. Click. Are you sure it's going to cost this number of hours? Let's do that. It's going to notify us in around 2 minutes while it's generating. Let's check out other features that's new to Mid Journey. Okay, here I have a realistic photograph of Burster creating later art in a cozy coffee shop. Wo Ambience, let's upscale an image, for example, this one. Okay, the image is upscaled. Now here you can see that we have way more options that we can do. The very region is the added in painting, which allows us to edit certain specific parts of the image. Let's do that. It's going to open a window here. Now we can actually change a specific part of the image. Let's highlight this. Now I want to say put hands holding a cup of coffee. Let's generate. Okay, now we got four images with the edited part. As you can see, the woman is holding a cup of coffee, but that cup is quite huge. I think the best one is this one. Let's zoom into the hands seems okay. But what I find about these journey edits is that sometimes they're quite out of place. For example, like this one, there is still a room for improvement for Journey in editing. Okay, that's the editing here, for example, let's choose number four. Now we can upscale and upscale four times. Basically it's just going to improve the resolution. Let's just do two X. Okay, while it's upscaling, we actually have our style tuner ready. We've got the link here. Here was our prompt and minimalistic logo of an Angry Cat. We can compare styles. As you can see, even 16 images is pretty good to make a decision about a style. Yeah, I liked the first, the third one here. We can choose it here and it's going to give us the code here. There is actually another method to choose between styles. We are choosing between two styles, this one or this one, or neither. If you don't like neither, another way is to pick your favorite from a grid. Here you get a big grid and you can choose between two images. This one for example, or this one. In this case, this image or this image in this big grid. It might be easier just to see different styles. It's up to you which one you use. At the end, the code will be the same. If you choose the same styles, for example, if I just choose this one, it's exactly the same as on the other method. Okay, here we can try to combine two styles. Let's say this one and that one. They are pretty similar, so let's copy it. Let's go back to at Journey. And let's imagine here we have a manual sic logo of an anger cat. Here we can change it to, for example, a dog. Let's try. Here are the images. There are slight differences between the logos here, but the overall simplicity and style is quite similar. This is how you can use tune command to generate many images for. Your specific prompt, choose the style in which you want your images. Now with this code, you can generate consistent style which was not previously possible before. Okay, now let's get back to upscaled version. Here we have the two x upscaled version. As you can see, the resolution is way higher here. I don't quite like this cup here. You know what we're going to do? We're going to go back to the original image, the one that was edited, and we'll upscale this image by a factor of two. Okay, here we've got our original image upscaled. What I wanted to show is that even though the resolution is higher, but the problem is the elements and details, it doesn't improve them while upscaling here we do have a problem with the hands here. It still persists in the upscaled by factor of two version. This is just something that you need to be aware of. The last feature that I want to show you is the weird parameter. Let's see what Jeri has to say about it. The optimal weird value is dependent on the prompt and requires experimentation. Try starting with smaller values such as 250 or 500, and then go up or down from there. If you want a generation to be conventionally attractive and weird, try mixing higher stylized values with weird. Try starting with similar values for both. For example, a cat stylized 250, weird 250, here are, this is the weird zero result and weird 1,000 Again, this is weird zero, looks pretty normal those two. Let's see, the weird 1,000 Okay, what's the difference between weird chaos and Stylize? Chaos controls how diverse the initial grid images are from each other. Stylize controls how strongly mid journey default aesthetic is applied. And weird controls how unusual an image is compared to previous mid journey images. This is something you just need to experiment with and we'll try something basic here. Let's imagine a cat in at, let's weird zero, then let's go common a cat in the hat. The 500 for the last one, let's make it crazy. 1,000 weirded. Okay, let's see, this is weird, zero, weird 500, the same as zero. Maybe the prompt is not the best one for trying the weird stuff, weird cat and had 1,000 Here we definitely see more interesting things going on on the hat as well as the cat is now a little bit different. Reminds me of a header from Alice in Wonderland, this one. These were the updated features in my journey. I especially like the tune command. So go ahead, play around and experiment. 53. Introduction to Basic AI Photo Editing Tools - bigjpg.com and vectorizer.ai: Hello. In this module as well as the next few modules, we will be covering photo editing AI platforms. So what's that? Well, before when we used mid journey Dali staple diffusion, we used text to image generator. So basically you provide a text and AI generates an image from that prompt. The photo editing, you can take that image and improve it. You can change the lighting, change the background, maybe cut out the image, or convert the image to a different format. All those things now we can do with AI as well. In this module, we will cover the simplest tools. These are very easy to use, they are for a specific purpose. For example, here have the biggbgt com that enlarges images without losing quality. Then there is vectorizer that converts PG and PNG images to SVG vectors that's useful when you want to create a logo. For example, you generated logo with Journey and you've got the p image. But now in order to make it into a logo, you actually would need to convert to SVG vector. This is where this app will be handy. Then we have the segment, Anything.com That's a research demo by Meta and that allows to cut out any object from the image. Another app that I've decided to include here is Creator.com This platform is mainly for e commerce purposes. It helps to place your product in a nice background. It helps to generate background for your product. Okay, let's start with the big.com here. When we go to the website, it's pretty basic. All you need to is click on the Select Images. And select Image here I'll go with the concept art that we've generated with Mid Journey. Then when you upload the image, all you need to do is to click the Start here. You can choose the image, type artwork, or photo. You can upscale up to four X for free. If you want a higher upscale, then you would need to upgrade for noise reduction. Noise reduction, basically it fixes a grainy area of the image and smoothens out the lines. If you find that your image has too much noise, then you can select more higher noise reduction. I think here let's select a medium. Let's click okay. Let's also select a few other images. Here I have a low resolution photo image. Let me show you, this is the low resolution photo image, because if you zoom in, you can see pixels and just Noise. I will upload this photo photo. I'll click Start here again. I can choose the image type. First I want to try the artwork, then I will also do the same, but with the photo you can see the differences noise reduction. Let's do the highest, and let's again use the same image. Now I also like to photo for X and highest. Let's check out these images here. The first image is the original one. It's the image that was generated by mid journey, and that's the highest quality that we could get from mid journey. We zoom in. As we zoom in, you can see that we start seeing those pixels and more noise when we scaled this image here is the upscaled version. If we zoom in here, you can see that the lines And overall the shapes are way more smoother. If you are trying to print a poster, then using a tool like this one that enlarges the image even further may be very useful because now it's nice, clear and sharp. However, one thing I want to say is that if there are any artifacts on a mid journey, for example, with these R or sleeves, then you would have to manually fix it. Because when you enlarge it, those things will be the same as the original image. If there are certain things that you don't like in the original image, then you would have to fix it yourself. Okay, let's move on to the photo. Here is the original photo image. As you can see, the quality is not great. You can see you have these grainy areas. Okay, If we compare this original image to the upscaled versions. This is version where I chose photo as the image type. Here I artwork as the image type. See, there is quite a difference when we choose the artwork type. Then it smoothened everything here. Now we don't see any noises, but just the whole photo now looks more like a drawing or painting, perhaps it doesn't have that photo realistic effect in the photo type here, we still have some noise. It didn't get from the overall noise, but it did improve the quality a little bit here. But overall, I like the artwork type better here. Okay, so this is the big Pg.com There are plenty of other tools that enlarge images, and they use different algorithms and stuff, so you can try finding the one you like the most and stick to it. Okay, let's move on to vectorizer. Here again, very simple interface. You just need to drag images. Let's try with some of our logos, for example, this one. It processes quite fast. This is the original image. If we zoom in a little bit, you can see that the original image gets the pig cells. But the vectorized result, it doesn't. You can see that the lines are nice and smooth. This is very important for logos because with vector image, you can make it into any image resolution you want. For example, if you want to make your signboard where it has to be a very big image, then this vector would work as well as a small menu image. That's what's great about the vectorized result. And you can download it very easily here. You can choose the Pig or EPS, these are both the vector formats. And then you can choose the version and other settings. And then just click Download. 54. Basic Photo Editing Tools - Segment-anything.com: Okay, let's move on to our next tool, the segment, Anything.com by Meta. Here, when we go to the website, we see this landing page with information about the model. If we want to use the tool, we need to click on the stride. The demo here, we will need to agree to the terms and conditions. Here it says that this is a research demo and may not be used for any commercial purpose. Any images uploaded will be used solely to demonstrate this segment. Anything model, all images and any data derived them will be deleted at the end of the session. Any images uploaded should not violate any intellectual property rights or Facebook community standards. We need to agree that here we have a large gallery with images. You can select and try this model on these images or you can upload your own image. Let's upload our own image here. For example, let's use some con we got from mid journey. Here we have con. Now the segment, anything will process our image. Here we've got our image. If we point to any element of the image, you can see that the whole element gets selected. For example, if we point to the ****, the whole **** is selected. If we point to the book, then the whole book is selected. Let's try that. I point, then I just need to click. Now the recon is selected and I can cut it out. Here I have the cut out object option. I just need to click this. It will cut out my ****. And as you can see, it just cutted out the portion of the ****. It didn't put the book or anything or any of the background, just the **** here. If I want to use this image, I can copy the image, or I can save the image. Let's save. And here is our image. Okay, what if I want the **** with the book? Let's go back here. I just need to select the **** first. So you can see this blue dot when you select something gets added to the element. And then I also want the book, so I'll click on the book. Now we've selected two elements. Here. Again, I can go to cut out object. Now we have this raccoon with the book. Let's save the image. Now we have the raccoon with the book. If you want to use that image anywhere you have this PNG image. Okay, In a similar way, I can add multiple elements. For example, add the ****. I can add the book. I can add the lamp. I can add the armchair, and so on. Maybe a window or something. If I don't want a specific element in the image. For example, it selected something that I didn't want, then I can click on this remove area and click on the element that I don't want to see. For example, I don't want to see this element here. As you can see, it removed it from the selection, or for example, I don't want to see this now, it removed that element. Okay, Another way you can select elements is by using the box. In the first one, we just hover and click on the element. Here, you can use the box around the element you want to see. For example, if I want a raccoon with the box, I just make the box around the raccoon. As you can see, again, only the **** is selected. If I want the book, then I need to click on it as well. I will need to click on the book. Now we have two items selected. I can cut out the object and it will be in my cut out gallery. Another thing you can do is Use Everything button that basically will scan your image for all the elements. If you want, you can cut out everything. You can cut out all objects. Now you can see that every element is cat out. We have the raccoon, we have the orange chair, we have strange objects and so on. I think segment anything model works great on product photos. For example, if you have a product photo with some background, maybe a black background, and you want to just have image of your product without any background, then you can use this tool that will be very handy. For example, let's try that. Here I have an image of sneakers that I've generated with mid journey. Here we have the sneakers. Let's say I want to just have the image of the sneakers again. I need to select one here. Let's select it. We've selected one, then let's select the other one. If you made a mistake for some reason, you can always undo or reset, or you can redo. For example, here we have selected our two sneakers. Let's cut out the object. Here we have it. Let's save this image. Now we have a very nice cut out image of our product. This is the tool that quickly allows you to cut out any object or element from your image before. What I was doing is, for example, this is the image I'm using. Preview. Preview. Here I have this magic wand and I was selecting it and trying to remove. However, with this, as you can see the selection, it just used the similar color. It's pretty difficult here to cut out it nicely. If you compare this to what meta cut out here, it's just in one click, you have the full object. Very easy. And a great improvement over previous tools. Okay, so that was segment anything.com 55. Basic Photo Editing Tools - Creatorkit.com: Now let's go to Create Kit.com The Created Kit is mostly for E Commerce and it's used to help you place your product in a nice background. Let's try it out. Okay. First you'll need to sign up or sign in. You can sign up with Shopify. I already have an account, I'll sign in. Okay, once you sign in, you'll have this button here created with AI. And here you can upload your product photo or you can try with one of these images. I think it's fun to try it out with your own product photo. Let's use the sneakers image that I've generated with Mid Journey, and then I made a cut out with the segment. Anything. And if you look here, it says that using transparent PNGs gives you a better result. Because otherwise they'll need to automatically remove the background. That may not be as good as the segment anything model. Let's use the PNG image that we have here. Okay, here you can make the image a little bit bigger or smaller. Sometimes that doesn't work and you may need to reload the website. That just helps you to choose a composition for your product. Okay, let's continue here in the settings. First, you'll choose your product category. What is your product? This is a footwear. This will help I place your product in a correct manner. Okay, Then you can choose styles. There's a bunch of different styles. Flowers, wooden floor, and so on. And also for example, here you can also see what it can be with the shoes. Here they have different examples of different products. Jackets and so on. Sofas, yeah, cool stuff. You can choose the style or you can choose to write your own prompt for the background. For example, here you have a chair inside and minimalistic studio with plants. Now we can write, you put sneakers. Sneakers in a beautiful mountain with snow can be something like that. The negative prompt is we don't want reflections. Maybe we can try that here. We can also select the model for created kit diffusion model, version 0.8 This is the model specific to the created kit. And here's also an option for the 1.0 but you'll have to have the enterprise plan to be able to use that maybe later. Let's try with the 0.8 version. Okay, this is pretty cool. I like the first one. If we go back here, I think this one is a little bit strange. The shoe was hanging in the air. But the other ones, yeah, it's all right. I think the first one is the best we can say image and check it out here. The background is quite nice. The only downfall here is that the image resolution is way worse than the original image that I gave. This is the image that I gave the app. If we zoom in here, you can see that our product is now pixilated. Okay, let's try something else. Let's go back and maybe this time choose something from the styles. I think I like the paint one. Let's see what other things they have. They have the forest, they have tulip field. The grading back drop also looks fun. Yeah, let's stick with the paint again. Let's select the footwear here, let's generate it. We've got some cool backgrounds here. I think number three is my favorite one here, just because it aligns with the shoes and we have a nice reflection shade here. Everything looks pretty good compared to the other ones where it's floating around. Okay, yeah, we can save that as well. Here we have this nice background. The only problem again, is that the resolution is pretty small here. If you want a higher resolution image, then you would probably use a different tool or different method. For example, creating the background in mid journey. However, what's nice about this tool is that it created those shades. If we go and see our original product, the original image is this one. As you can see, there are no reflections. And here the created it nicely added those reflections and integrated the product into our background. Okay, in this module we've covered basic and simple tools that help to edit your images. For example, we've covered the big GPG, which helps to enlarge your images. The vectorizer that converts your images to vectors then segments, anything that cuts out your images, and the created kit that helps to create a background for your product. 56. ClipDrop Introduction: Hello? Hello. In this module we will talk about Clip Drop. Clip Drop is a fun platform. It's an ecosystem of apps, plug ins, and resources for creators powered by artificial intelligence. Basically, it's an AI image generator, as well as a place where you can edit those images or any image or photo. It's developed by stability. We've talked about stability and we've talked another platform that they've created, which is Dream Studio compared to Dream Studio. In my opinion, Clip Drop has a way better interface. Clip Drop has a bunch of tools, for example, clean up, remove background, and so on. Let me show you the first tool we have here is the stable diffusion model. And that's the basic text to image generator. If we click on this here, you write your prout and you will generate images. But what's fun about the clip drop is that here you can try out the new models that stability AI develops that may not be released to the public yet. Similar to Dream Studio where you have the access to the newest models, here you can try it out. Okay, the next tool is the crop and that's the basic out painting tool. Then we have the reimagine that allows you to do variations of your image. Then we have the clean up that allows you to remove objects from your image. We can also remove the background, We can change the lighting of the image. It's especially nice, four photographs that we have the image upscale. We can upscale 2x4x in seconds. Then we can also replace the background. In the previous module we talked about create a T e commerce website where you can use your product and then change the background here. You can do exactly the same thing. You can write a prompt to generate background for your product or any other image. Then we have the text remover. If you want to remove text from your image, you can also do that. 57. ClipDrop Tools Overview - Stable Diffusion, Uncrop and Reimagine XL: Let's get started and let's explore those tools in a little bit more details. The first one is the Staple Diffusion. Here, I'm not going to spend too much time because we've already talked a lot about staple diffusion and we've tried the dream studio, and it's basically the same thing here. Maybe we can generate a few prompts and then move on. For example, let's put a fairy inside a purple galaxy bottle, Magical background here. We cannot put the negative prompt, but I believe we can choose the style here. If you click to a No style here, we can choose the style that we want our images in. Again, you can choose origami line art and so on. But for now let's just keep the default and let's click Generate. You can actually subscribe to have faster generations or you can skip. Okay. Not bad. We've got pretty good images. We've got this fairy in a bottle and she's holding another bottle that's pretty cool. With the free medium version, you have this water mark with the clip drop again. Well, this one is nice. You have some butterflies as well. If you want to subscribe to clip drop, the prices are quite affordable. If we go to pricing here they have the free version and the pro version, if you have the monthly subscription. Here you have the unlimited tools that you can use for stable diffusion. You have up to 1,500 images per day, which is a lot. Here again, we've got these beautiful images based on the simple product we've provided. I'm actually surprised because the quality is pretty good as well as the proportions. Again, this is, it should be a human like figure, not bad. Okay, let's move on to other tools. The next tool is crop, which is basically the out painting here. You can try it with their examples, you can check the original one was just this image when you extended the image. Now you have this, you can see with other things, this is the texture landscape, pretty cool. Let's try it out with our images here. I have a fun image, we can use memes here. I think it's going to be fun for out painting. I have this meme here now. I can choose which way I want to extend the image. I can extend it as a landscape. This is the custom one. I can write my own dimensions here. Or you can choose the landscape or portrait, or square. Let's choose the portrait. And we can move these as well. For some reason I cannot move the upper one or the bottom one. Okay? But let's try with this next. Now you can see that it generated pretty fast and it extended our image. Now we have four different versions and we can see which one is the better one. This one is pretty nice. She's in the dress, although what does she have here? The upper one is not quite bright. Okay, here again, we've got the space here. Okay, let's generate again and see if there's anything better. Okay, we're still getting something strange on the top here. The bottom half looks nice, but the top is just horrendous. Okay, I think the best image here is the second one, the dress. It's pretty good how it extended to the legs. Maybe this leg is too skinny, but overall, I think it did a pretty good job. Okay, let's try with something else. Maybe let's use another image. For example, let's use baby here. Let's again do the portrait. It's definitely has some weird things going on on top of the image. I'm not sure why the other images we have, this one is pretty good. It extended the C and now we have this nice Sky. The boy is wearing a cool shirt here, also changed the shirt and so on. Okay, so these are the things that you can do with the out painting in clip drop or it's called crop. Okay, Now let's move on to Reimagine here. You can again try with examples below. For example, let's try with the bed bedroom. Here we've got some beautiful interior design. This is the original image. Now we got some variations here. All of them. I like the colors here. This is interesting. The lamp. Okay, this was with the example image. Let's try out our own image here. I've generated an with mid journey. It's an image of the office. Let's see what variations we can make here. Now we've got our images here. This is the original image. These are the variations. This one is overall quite good. This one has a few artifacts on the floor. I think something wrong with the lamp here and the computers here. The original image that I've appl image generated by mid journey. I think this image is superior over the variations. The variations were done with stable diffusion. It's a journey. Still is a little bit better because here you have less artifacts compared to other images here. But overall the style and design is pretty spot on. No complaints here, especially that you can try this for free. Okay, let's go back. This is the re, imagine you can create multiple variations from a single image. 58. ClipDrop Tools Overview - Cleanup & Remove Background: Now let's move on to our next tool, the clean up. I think this one is pretty good. Here we can again try with some examples. Let's use this image here. Let's say we want to remove a pen here, we can just highlight the object that we want to remove. Then we can just put the clean. Now, it's really cool for brush size, we can make it smaller or larger depending on your image and what you want to do. Yeah, let's now upload your own image here. I want to use the image that we've generated with mid journey. As you can see, it has a lot of elements here. I'm just curious how it's going to remove things. Although the quality is not the best, I'm not sure if it's going to do anything. If you make a mistake, you can always undo it. If you want to move the image, then you can press the smooth button and move your image. I don't want to see the sculpture here or I don't want to see this person here, which will be even more intriguing. Remove that person from here. Okay, let's clean it. Okay, that's interesting. It just left this white space here. Let's try with a bit more realistic image. For example, let's use our me here. Let's say we want to remove the so here, let's do that. Let's increase our brush size a little bit here. Now I can remove the lady here. Okay, here, it's not perfect. That's probably because our image was too big. It covered a lot of the background, didn't quite capture the background here. Let's try it again. Let's undo it. Try to remove all this area, maybe that will help. Let's clean again. Here, we just have smudged area and that's probably because this image was too large. Okay, let's try a different image, So maybe we have a smaller things that we remove. Maybe that will work better. Okay, let's go back. Let's choose a different image. Here I have a photo of the beach with people. So here we have a few people here. And see if we can remove some of the people here. Let's maybe this girl here first, kill it clean. Okay, actually, not bad. The only problem here, we've got the reflection now. Let's also try to remove the reflection here. Okay, this is pretty good. Like nothing was there in the first place. Let's try a few others and see if we can remove all of the people here. Very good. Let's try a few more. Let's see if we can remove two at once. Okay, that worked. Let's remove everyone else from here. And boy, oh wow, we removed all the people here. And it's undistinguishable that there was anyone in the first place here. That's pretty cool. Now, let's remove those as well. I'll move my image then again, select also this boat. Let's remove that one as well. Let's clean. And maybe not sure what this is, let's clean this. Now here with this clean up tool, we've made a beach full of people to a secluded beach with no trace of a person. This was the original image here, we've made it from here to this. That's what the cleanup tool will do. If you have anything that you want to remove from images, then you can use this clean up tool. As long as the item that you want to remove is quite small and doesn't take a lot of the image, then it will work well. However, if it's too big then you might run into this smudge area thing. Okay, let's move on to our next tool, it's the remove background tool. In the previous module, we've talked about the segment, Anything by Meta that allows you to cut out a specific element. Here, it does pretty much very similar thing. It just cuts out your object from the background. Here you can read some more information. They claim that they have the most accurate background removal solution available and you can see an example. So this is the image here. You have the person's hair. As you can see, that hair was kept when the background was removed. However, competitors, usually some of the small hairs would be deleted. Let's see other objects. We have complex objects, we have this thread, Competitors have the left behind. Then some sharp edges. Clip drop removes image backgrounds and keeps the edges of objects extremely sharp here. If I zoom in a little bit here, you can see this sharp curve here. For competitors, we see some of the background left behind. The last one is focused only on the main object. Here, you can see that it can detect the object very well here. For example, here it knows where the stool is versus the shade. However, for competitors, the shades may be interpreted as part of the image. Let's try with some example image. Here we have the photo of a woman with lots of hair. So see how it's going to be removed. Okay, here we can actually move our background left and right here. If we move it completely to the left, we have this image without the background. One concern here, I think it left a little bit of the background here. Maybe it tried to save the small hairs, but now we've got lots of this yellowish colors here. It would need extra editing here. We can download this image. Actually let's download and see. As you can see, there's a lot of this yellow hue here. Let's try with a different image. For example, let's use the image that we've generated with mid journey. We have this glass with galaxy inside of it. Let's see, Overall, I think it removes the background pretty well. You can see that it's sharp and no, it doesn't leave any artifacts. And it's also very fast. We can download that here we have nice sharp edges as they promised. Good. Okay, let's do one more. And here I want to use the sneakers that I've tried with a segment, Anything by. And see how will the clip the same job. Okay, let's download this now. Let's compare to our, the image on the right. We got with segment anything by the image on the left. It's by, as you can see, the results are comparable. If I zoom in pretty much the same results, actually maybe the segment anything here is a little bit better because it didn't include this blue line here. But my conclusion that overall they're quite comparable tools you can use whichever you like the most. 59. ClipDrop Tools Overview - Relight: The next tool we can try is Re light here. We can choose image from example. Let's use this photo for example. Here by default you're getting two different lights. You can add more lights if you want with the new light button for the background. You can keep the original background, you can the background have more lighting received, light that is going to be affected by these lights. Or you can make it transparent. If we just click transparent, then we only work with the subject, with this person here. Then we have ambient. Ambient is basically exposure of the cut out of this person here. We can increase exposure or decrease it. Let's keep it, it wouldn't affect the background. You keep the background. Let's remove the transparent and keep the original here. It only affects this person here. See, we're changing the ambient light without any effect on the background. Okay, let's keep the transparent ambient somewhere in the middle. Then for lights, we can move out the lights depending on what you want to create. Then you can also increase exposure of the light and the distance, how much it can reach. Here's small reach versus large reach. Then we also have the radius. If you decrease the radius, that will also quite similar to distance, would decrease the light exposure. Here you also have the second light. You can try out playing with the second light, the first one, depending on what you try to create. Let's try with a different image. Let's use maybe a portrait. Here we have the portrait of girl with pearl earring, as you can see. Here we have the pre selected color of the light. Here you can actually choose a different color if you want. You can select maybe gray green. You can try different colors. Or if you have a specific color, you can input that color here. For example, I think maybe brownish color would work well here. Maybe like this on the color feels more natural here. Then the second light is blue. And then we can manipulate the slight to make a visual effect on this image. For example, here, be nice. Okay? Or you can delete this light if you want. Then you were just left with light, one source of light. Okay, let's keep the second one. I just want to make it blue to match maybe her scar. Now we can make the exposure a little bit less. Yeah. And if you want, you can add multiple lights again. We have different light here, make exposure smaller up to your image nation. When you're done, you can download the image. Let's download this image. Here we have it. Another thing I want to try is I want to try it with a product image. Here we have the image of the product, I think with the product photo, this is actually a really good tool because you can spend so much money for a professional product photo shoot Here you can make it in one click for the background. I've actually uploaded the PNG, but you actually a background. You can add a background if you want. You can add a specific color, for example, white. You can also make it to receive light. Now you can create this fun effect for the background. Maybe we can make the ambient a little bit lighter. Here we have the light. Let's make the green exposure higher. Let's make it green, Let's make it blue. Well, this is worse. But basically here, you just need to play around with these different settings until you find the perfect combination. So I'll try to do that here after playing around with different lights. Now I've added one more light. Light Number four, the background, I used a lighter color. It wasn't too black because black wouldn't receive any light for ambient. I made it a zero. I don't want too much exposure for the lights. We have some distance here, not too much then the distance is similar for all these lights. It really depends on lighting and where you want it to be, just changing settings based on your product. Okay, after you're satisfied with the image, you can download it. I'm happy with this one. Let's download it and see the final result. This is the original image and here's the image with the lights. I really like this one. I like how it highlighted the night sky on the shoes. Pretty cool. That's the real light tool that you can use in clip drop A. 60. ClipDrop Tools Overview - - Image Upscaler: Let's move on to our next to image scalar. Here you can upload your image or photo and upscale it to 2x4x8 x and 16 x. However, if you want to upscale more than two X, you will need to pay for subscription. As you can see, you can subscribe to an annual or monthly plan. Okay, let's try with some examples here. This is, I assume the two X. This is the original image, as you can see, it's grainy image. Some things are blurry and pixelated When we move to upscaled version here, let's see, on the right hand side is the upscaled version. Here you can see the skin texture very clearly overall, it's very high quality. On the right hand side, the upscaled version, okay, this is the photo realistic image. Let's see some other things, maybe with the objects. Okay, so we've got this blurry, pixelated image with upscale. We got the C sharp edges. Very good. Very good. Okay, let's try with our own images and see how that compares to the tools that we've tried before. Okay, I will upload the same image as I upload it with Ppg.com Okay, let's scale. Let's check it out. This is the original image here. Now the app scaled version. Let's move in a little bit. It's definitely cleared some areas, made some smoother lines and outlines. But I would say that for this example, I liked the result with Big GPG better. Let me show you what we've got with Big GPG that, you know, this is the image we've got with Big.com and that was four X. Even though this is two x, of course the quality will be less than the four x. However, in my opinion, the overall I really like the way the big improved this image here. Okay, let's move on to a different image. Okay, here I will choose the photo that we've used. I've used image here. Let's upscale it. Okay, let's check it out. Here is the original image when Drug. Now it's a two X upscaled version. As you can see, this is better. The upscale worked quite well here. Here on the left hand side, we have this gray image. The upscale smooth the photo out very nicely. Let's see, even with two X, it did a pretty good job here. Okay, let's check what's the difference with the big here. Here on the right hand side, we have the image up scaled by clip drop with two X. On the left hand side we have image upscaled biggs. As you can see, I think the clip job did a better job here in terms of it removed the patchy area. If we look at the hair for example, we still have those patchy areas here. It's way more smoother. We can see that the hairs are more clear and distinct. If we look at the sleeve here, look at how sharp is the edge here? Here? It's a little bit more blurry. Overall, I think with this photo clip drop did a better job. Okay, I wanted to try the 16 X and detailed version, so I bought the subscription. Let's try our photo with the 16 X and see if there are any differences. I'll choose detailed and 16, once you have a pro version, you'll have all your images that you've generated here in the history. You have up to 14 days to save them, okay? Let's check out this image, okay? Here, the detailed up scalar actually made it more blurry. So this is the original, we've got these blur clusters, okay? So that was the detailed, the detailed setting. And 16 x, I've actually generated the 16 x with the smooth, let's check that out as well. This was the smooth up scalar. As you can see, this is way better than the original. It removed the noise and enhanced the image. I wouldn't say the 16 X is too different to the two X because here we still have some areas for improvement in terms of the colors and the skin texture. But it's probably because the original image was just so poor quality that there's not much to work with. Okay, in my opinion, the two X version was very similar to 16 X. Here in this particular photo clip drop may be a very useful tool if you want to upscale and nose you images. Here you have a smooth or detailed version. If you have a subscription, you can try to do four x or even 16 x, depending on what you want to achieve with your image. One thing I want to point out here is that the image up scalar is a bit different to photo restoration if you are looking to restore your old photo, for example, old photo like something from the 20th century, black and white photos and so on with scratches and so on. Then image up scalar may not be the tool that you're looking for. You may need to Google some old photo restoration AI tools right now. There's plenty of them. They will restore your photo. For example, I just found a random one vans AI. As you can see, it allows to restore this old photo and completely removing those scratches and those folds to make a colorful image. Here, those are the tools that restore the photo. This is the tool that scales even though it can remove the noise and enhance your image. It's not a photo restoration tool, it's the upscale tool. 61. ClipDrop Tools Overview - Replace Background: Let's move on to our next tool. It's the replace background tool. That only works if you have the subscription. You need to buy the subscription to be able to use the tool. Here you can try some examples. Here you can see the photographers. Here is the original image. Now we've replaced the background. You can see the cut out of this person. What's noticeable is that the hair, you can see the cut out because the a lot of hair, it's not as smooth. You would need some photo editing to improve the quality. Then we have in the park, we removed the background here in all of these three images. For me, it's apparent that this was a cut out just because of the lighting and the background doesn't match. These things would need improvement. Let's check other examples. Creative agency, here's the car, then we've changed the background here. I think for the object, it's a little bit better. For example, for number three, it's more realistic here. Here at least it added the shades to make the object fit in the background. Let's try some other examples. This is the sneaker to remove the background. I think with the shoe here, it actually did the best job. Selfie, I think would be the same as with the person that we've seen. That yeah, the cut out is quite apparent here. Okay, the conclusion here is that for objects it works way better than for people. For now, let's try with some objects here. In the last model we've talked about creator kit that allowed us to replace background, we're giving an image of our product and then we were choosing a style or we were writing a prompt to generate the background here we can see how the clip drop compares To create a kit. For that reason, I'll be using the same sneakers. He our PNG image. Let me upload it. Okay, here are our sneakers. Let's use some prompt. Maybe modern colors, modern colors, splash ground. Let's click Generate. Okay, this is actually pretty fine. We've got some sparkles here. Here are the four images that we've got with this prompt. I think number four and number two are the best ones here. Let's try a different prompt. Let's use the sneakers again. Here we've got a random background prompt. We can use this button to randomize it and see if we like it. In this case, I want to compare this to Create. I will use the same thing that we've used there. Let's use our sneakers on top of the mountain. Here, I'll put sneakers placed on the Snowy mountain. Let's generate that. Okay. Here we see the mountain background and we can actually see that sneakers are sitting on the Brock or Snow. Let's see other images here. Here we have a good integration of sneakers with the background. Because here, the shades. Yeah, I think this one is very nice. Here again, we've got these shades that make it more natural. I think the best one here is the second one. Let's download it. Okay. On the left hand side we have the background generated with clip drop. On the right hand side we have the background generated with created T. As you can see, both platforms integrated the product, the Seekers pretty well. Into the setting. It added the shades here as well as here. Both platforms did a good job on that. However, the big difference here is the resolution of the image. Here we've got a much higher resolution compared to the created kit. This is the image that we've got. I didn't minimize it or anything as you can see. If we zoom in, let's zoom into this image. Image of the product is very poor quality area here, it's all pixelated. If we zoom here, it's still a pretty good quality if we compare this to the original image that we've supplied. So that was the original image zoom. Okay, Maybe here it's started to get more pixilated. The clip drop has shrinked the product image a little bit. If we zoom in here, it's worse quality. But overall compared to create a kid is way better. So now we have this nice big image that you can use on any platform. For example, if you want to post that on Amazon, that would be a good enough image to post that compared to this small image. In this case, I would use the clip drop because it makes a way better quality images and that's what matters. Okay, let's try with the same tool. Let's try some different format. Here I have the image that we've generated with Mid Journey. Let me show you. I have this girl here. You can see this is a water color image. I want to see if this replace background tool will adjust the background to the style of the image that I'm providing. Okay, let's try that here. We can make it larger or we can rotate it. Let's keep it like that. Okay, the image I've provided, it's not a PG. It has this white background. What it did here, use the remove background tool to remove the background first. If you don't like the way it removed the background, then you can provide the PNG image here. It did it automatically and as you can see, it removed it removed the toys eyes because it was the same color as the background. Let's just try this here. I can use this randomizer. Let's just put here a beautiful landscape, and let's click Generate. Okay, Here you can see that the style was adjusted to the image. We didn't get any photorealistic background. Now it's more watercolor, blurry and so on here. I think the number two is the best here. All of that looks watercolors. Let's see, the first one here, we've got different style. I think this looks more pastoral style. Doesn't match too well with this image, but the number two worked great. Even though in the prompt, we didn't put anything, we didn't add the water color, we didn't add any style. But it just used the style of the image to make a background that matches well with this image. I think that's a very smart tool. Okay, we can download this. As you can see, this replaced background tool will allow you to place your characters into different settings and background. This is the replaced background tool that allows you to generate a new background for your product image or for a character or a specific object. 62. ClipDrop Tools Overview - Text Remover: Let's move on to the next tool. The next tool is our last tool for clip drop is the text remover. Okay, let's see the use cases. Here we have the creative agencies. Let's see. This is the, the original image and you can see that there are those letters. Now we've removed them. Okay, let's try the molk. Here is the product. We don't want to see any text. We remove the text and you can see that there is the light logo. It didn't remove the slight logo, it just removed the text image editing. Again, here we have the slogan, we don't want the slogan. We remove it and we can put any other message here. Okay, great, here on the T shirt. That also removes the text. As you can see in all those four scenarios, it was able to detect the text and remove it. Okay, let's try with our images. Here with me Journey. We've generated some logo images because the text was legible, but we would want to replace it with something else. That's why this tool can be very useful here. As you can see, we had the text here and now it was removed. Now we can download now. We can put our own name off the bakery if we want to. This was the logo with the text. Now it's removed and we can add anything here. Let's try a different one. Let's use a different logo here. We have this text here, which is relatively easy to remove. Let's try another one. Here we have more text. Okay, this is interesting. Let's go to the original image here. This is what we've generated with mid journey. And you can see that this bit wasn't removed, possibly it wasn't recognized as the text. We can try again and see. Maybe if we try it again, then it would be removed. Let's see. Okay. Yeah, it didn't remove it. It didn't recognize that as part of the text. Another interesting thing is that it changed this cupcake base. It made it worse. I don't know if it recognized the text and wanted to remove it, but clearly there were no text here. It would be easier to remove the text ourselves here than rely on this AI because now we have this messed up cupcake base here. I'm not sure why this happened, but as you can see on other images, it worked really well. But this one, for some reason, it didn't give us good results. You can try it with your own image or any other image where you want to remove your text. Another thing that I want to try here is uploading a photo of a product with a lot of text and see how well it would remove the text from it, the same as with this more cup. Let's see here. I found two images on the Internet. Two photos. I wonder how well the text would be removed from this product here. Okay, let's see. Okay, not too good of a job. Let me show you the original image here. This is the original image. Tried to remove the text, however, it created some smudge area. I don't even know what this is. Maybe try to replace it with bubbles or something. But overall here, because it's a bit more difficult, because this label is transparent and we have this gradient in the background, it's more difficult to remove the text from here here, It didn't do a clean job, actually. Maybe the clean up tool would work better here. Let's actually quickly go and check it out. Okay, this is the cleanup tool. Let's upload our image. Okay. And here, let's try to remove this text text here. Okay, let's clean this up. As you can see here, we've got a way better result than the text removing tool. Sometimes the clean up tool would be more appropriate than the text removing one. Maybe this part doesn't do it quite well. Actually, I'm curious what's going to happen with the high quality. Let's go back. We have those text selected. Let's choose the high quality. The high quality mode takes a little more time to reprocess your image for better results. Okay, let's use that. That's only available for people with subscription. I think that was quite similar. I don't see too many difference with the previous result. Okay, so as you can see, the clean up tool actually here in this case was better than the text removing one. Okay, let's go back to our text removing tool here. I want to use the other image. It's also the image that I found on Internet and as you can see it has some text here. Okay, again, we didn't get quite a good job here. Unlike what we've been promised in the use cases with a mocap where it removes very cleanly the text here, when we applaud our own images, you can see that there are some problems still. Then for this case, you can try to use the clean up tool. Okay, now we've discussed all the tools that Clip Draw has. These tools can be very useful and helpful when you do photo editing or image editing. Here they even describe some use cases for these tools. For example, team portraits to automatically harmonize your team photos to create a beautiful and consistent team pages. Car resellers, you can place car in a nice background or real estate. For real estate, you can improve the quality of the image. You can upscale, you can change the lighting, for example, and also e commerce. Because there are a lot of tools that allow to make your product photo better, Remove background clean up, and all those tools will help to make your image better for me. My favorite tools in Clip Drop are light and the clean up. I think the is quite unique to Clip Drop because I couldn't find anything similar to this in any other platform. And what's also really nice with Clip Drop that there is a premium version that allows you to try it out and if you like then you can subscribe and pay membership. Okay, so that's it for the clip drop. In the next module we'll move on to Adobe Firefly. 63. Adobe Firefly Introduction: Hello everyone. In this module, we will go on exploring Adobe Firefly. Adobe Firefly is AI image generating and editing app. It was developed by Adobe and it launched quite recently in March 2023 and became available to everyone only in June 2023. Firefly has a text to image generator and it uses its own model. It's different to the stable diffusion. It's not based on stable diffusion, it's their own model and it was trained on a dataset of Adobe stock. We can read more information from their own website. Here they have the article, how Adobe Firefly is different to stable diffusion. So if we go here, how Adobe Firefly differs from stable diffusion. And here it says that Adobe Firefly is a family of creative generative AI models currently in beta plan to appear in Adobe Creative Cloud products including Adobe Express, Photoshop and Illustrator. It was trained on a database of expired copy right, and openly licensed images. Firefly takes text descriptions and translates them into AI creations. From amazing images to unique text effects. With more models planned. Adobe Firefly basically uses their own proprietary model, which is different to stable diffusion. Here they don't go into details for the differences between stable diffusion and Adobe Firefly, but we have this general understanding of their models and how they're trained. If we go a little bit down here, what tent does Adobe Firefly train on? And here it says Adobe Firefly trains on Adobe stock imagery in accordance with the stock contributor license agreement. Openly license content in public domain content with an expired copyright. This means anything you create in Firefly will not infringe on the copyright of artists. It also opens the possibility that everything Firefly generates will eventually be suitable for commercial use. So I think this can be very important for some people and why someone may choose using Firefly over other models, for example. Okay, what tools do they have? Well, first of all, they have their own text to image generator with their own model. Let's go and check that out. This is the text to image tool. Then we have generative fill, which is basically in painting. If you go here, generative fill, you can use a brush to remove objects or paint in new ones. From text descriptions, remove objects. In the previous module, we covered the clip drop. They had the tool called clean up. That's basically the same thing. You can remove objects here. Also, apart from just cleaning up the image or certain elements, you can paint in new ones. From text descriptions, that's basically in painting. You remove certain part, you write your prompt and it will generate something else on that spot. Then we have the text effects. This is a unique tool that Adobe Firefly provides. You can create custom cool text effects with that like feathers, bread balloon texture. We have the scales and so on. Then we have generative recolor. It will generate color variations of your vector artwork from a detailed text description. Here you need to provide a vector artwork like SVG. It will help you create color variations from your image. It will recolor, make a different color palette. Here you have this yellow and now it repainted orange. Then we have tools that are not publicly available yet. They are in exploration, as it says here. You cannot use it yet. I will not be covering these tools, But let's just check them out and see what does Firefly plan for the future. This is the three D two image. It creates a three D scene and uses a text prompt to generate an image. Here you provide a three D model. From that three D model, you write a prompt and can generate images with that, which is pretty cool. Then we have extend image. That's basically out painting. You provide an image and then the tool allows you to extend the boundaries. Then if we scroll down here, there's a bunch of other tools. We have the personalized results. It generates images based on your own object or style. I think this is very useful, especially if you're working with specific style and you want to create in that same style. Right now, it's a bit challenging to do that. And I believe that Adobe Firefly plans to make that available and allow you to do that quickly. Then we have this text to vector. It generates editable vectors from a detailed text description. I think this is very innovative because what right now we have is text to image, we get the Pec image. If we want to make a vector out of that, we have to convert that Pec image and make SVG or other vector format using other tools like we tried using Vectorize to convert our image to vector. If we're making logo, how wonderful it would be to just use this tool, write our prompt for a logo, and create a vector right away that will save a lot of time and especially that you can edit it right away. That also will be very helpful. Then we have this text to pattern that allows to generate seamlessly tiling patterns for a detailed text description. Then we have text to brush. Here we have new models that allow us to create something else from the prompt, not only image the Pac image, but here we can create vector. We can create patent or even brush that can be used in Photoshop to add to our own designs or artwork. I think that's pretty cool. Then we have the sketch to image, you provide your sketch, and I will make an actual image out of that. There are actually already platforms that allow to do that. This is nothing new, but it would be fun to see it on Adobe Firefly. And then we have text to template. It generates editable template from a detailed text description. Okay, That will be very helpful when you do a website design. You can write your prompt and immediately be able to change the text and manipulate this template the way you want, which is amazing. Okay, so here are some future tools that hopefully will be available to us. Okay, but right now we have these four tools that are also very interesting. We have the text to image, generative fail text effects and generative pre color. That will go on and try these tools. 64. Adobe Firefly Text to Image - Portrait & Logo: Let's begin by actually checking out what kind of images we can generate with firefly. And for that we can go to gallery and see some of the successful images that were generated by the community. Okay, so here we have some text effects. Some three D renders, water color paintings. We have some faces here. This one looks very realistic here. Okay, I, overall, I can see that the images are pretty good. Another thing that I'm noticing is that the prompt is very short in order, for example a key origami, rose, flower holographic, close up digital in order to get a good image. It doesn't require you to have a long prompt. Unlike stable diffusion, let's say let's try to generate some images ourselves here. If we go back and click on this text image generator, let's generate images here. Basically, there are no settings here. We can paste our prompt here and just click Generate. Let's do that. I'll be using the same prompt as usually. Okay, we have this professional portrait photograph of young British woman and all those stylizI'm. Sure if we need those styles here, but let's keep them. Let's see, Click Generate. As you can see, it automatically placed it in the art category, the Quantum type. It placed it in art. Let's check those images. Okay, the images are not grad. Actually, if we go to the fireflies gallery, I didn't find a lot of portraits probably. That's why maybe portrait is a limitation, as we can see here, compared to mid journey or other stable diffusion models. That's definitely a worse quality here. Okay, can we make it better? Let's try it out in terms of quantum type instead of art. I'll choose photo here. You can choose different styles here. There's quite a few synth wave science fiction. I've seen that the pop art image was pretty good, but it's not going to give us the photo realistic style that we're looking for. Okay, I'll just remove some of those, stylist, keep it very simple. Blurry, rainy city background. Let's generate. I'm not going to choose science fiction here. You can choose the style. You can also choose color and tone, lighting and composition. For color and tone, we have those different colors. We have black and white, warm tone, cool tone, pastel color, and so on. For lighting, we have some options from back lighting, dramatic lighting, low lighting, studio lighting here, composition, blurry background, close up, white angle, and so on. Here we can choose blurry background. We can choose close up. So here you can only choose one category, because I already say blurry background. I don't see why should we add that again. Let's choose close up. Let's try this. Let's, let's see. This image is not too bad. Well, actually it has very natural expression here and facial features. I don't see any artifacts. However, the other ones here, we definitely, especially like hairs here, there is a problem here. And the other one, I think here, her face is a bit distorted just we have this cut here overall, the only good image was this one. I can download it when I open it, this is my image. It has this Adobe firefly, just so you know. Let's try with other prompts here. Let's go back this time. I want to use Logo here. From the images I saw from the gallery. I think logo would be very successful here. We have really great images of illustrations and geometric, geometric images here. I think logo would be really good here. Let's try it out. I have the line logo of Cup Ce with a tear and top clean line, simple shape, minimalist vector. Let's generate again, it shows the art style we want the graphic, I'll change it to graphic now. Here I want to change the style. Here we have some effects like iridescent, dark isometric materials, clay, origami. We even have line, you can treat the art cartoon vector. Look, here we have a line drawing. I'll choose the line drawing here for lighting and composition. I'm not going to change that, but I will change the color in tone. Here, I want pastel colors. Okay. Now let's click Generate. Okay. Here, if you noticed that the images, these images are very similar to our previous images. If I go back and forward, there may be some new elements, but overall they are almost the same images, y here, if we go back, I've changed here the style, I've changed the color and tone. And that's it. If you change the style, color and tone, lighting or composition, or even add something to your prompt. Let's add something like color. See how the button changed from Refresh to January. If you go back, it was refresh. If I modify, my prompt will add style, color, lighting, or composition. If I add color, then see how it changed the button when the button is generated, then we will only make variation of these images. It's going to be very similar image, but some elements of it can change color, light, whatever you want to change will change on that image, but the overall decomposition will stay the same. That's when it's generate. However, go back here. If you click refresh, then it's going to give you a totally new badge of images. Let's click on that. Now we've got a completely new batch of images. Now if we change the style, let's put our line drawing, then our color Pastor color. Let's click Generate. We'll have similar images, just the variation of them. This was the previous badge, these are the variations of that. If you don't like these images, then you need to click refresh, and try new images. However, let's say you like a specific image, but you don't like lighting or you don't like the color choice, then you just need to change the color to whatever you want, like warm tone. Now it's going to give you the variation of that image, which is quite similar to using seed in stable diffusion. Let's actually refresh and I'll change that to pastel colors. Okay, This one is not pad actually change maybe the lighting a little bit. Golden hour, let's generate, it's too yellowish. Okay. Then let's remove the golden hour, keep the pastel color. Let's generate again. Okay, this is a little bit better. Let's try our other logo. We have a tree inside of a water droplet. I'll remove the line drawing and I'll paste my prompt here. And I will remove some details because it might confuse I, contemporary style. I'll just keep it simple tree inside a water drop blood. Let's keep it with spectrographic, one color white background. Okay, let's generate. Okay, this is nice but it has too many details. I want it to be very simple. Let's maybe add simplistic here. In styles, there's also a style called minimalism, and also we can try geometric because that might make the shapes more simple. Okay, let's generate that. Okay, It is now getting a bit more simplistic. I'll play around with a little bit more and then show you the final results after trying a lot of different styles. With this prompt, I found that the simple wire frame, as well as combining that with the golden hour lighting, gave me a really interesting results. I like the color scheme of these images. I've got like this gold color that the golden hour adds to the image and the wire frame gives this intricate details of the image. If you're making illustration, then maybe wireframe would be nice to use. However, the best style for this particular prompt, in my opinion, was not the wire frame but the geometric. When I choose geometric, then I get more simple logos. Let's check this out. Once I put the geometric, now we're getting those very simple shapes that I was looking for. Okay, again, we have this golden hour. We don't have to have it here, but let's just create a few more images for you to see. I think number one is very good and with a little editing, we can make a nice logo out of this. Let's save this. The other images are also quite simple. I like that all of them have simple background. It just takes some patience to try all those different styles until you get the image that you want. In my opinion, the firefly did a way better job with this prompt than even compared to mid journey. Because here, the model that firefly uses, it was trained on more illustrative images like these ones. Like logos, illustrations, geometric images. That's why creating specific images like logos may be easier with this firefly model than compared to other models. Because here you have all the resources to choose what kind of illustrations you want, what kind of logos you want, and you have all those different themes that help you get the art that you want. Although I think the limitation here is that yet you cannot use a specific image for reference in the future. As we've talked about all those new tools that Firefight may implement, then you can train your own style images and generate in the style that you want. 65. Adobe Firefly Text to Image - Illustration, Anime, Landscape and Concept Art: Now let's move on to our next prompt. Three D render, we have our three render of raccoon. Let's see. What I'm noticing is that these images have a lot of texture. Look at the ***** fur and then we have the armchair folds the carpet of those small details. However, I think there's some artifacts with the Pow overall. It's not bad, but it still needs some improvement. Okay, that was the art, let's check out some popular styles. Digital art might work here. And then for color and tone, let's choose warm tone lighting. Let's low lighting. I think it was going to go very well with this night lamp and composition. For now, I will not choose composition here, low lighting and let's generate. I would say that firefly is still not at the level of creating complicated three D renders of animals. Well, some of the images that we saw in the gallery were great, but that's not happening here with the raccoon. Okay, in my opinion, mid journey was way better. Okay, let's move on to the next prompt. It's the illustration. Hopefully here it's going to perform very well. We have the children's book illustration here. I will change the style, make it a graphic style, and let's remove the low lighting and warm tone. Okay, I don't think it captured our artist here because this seemed to be a very different style to the one I've indicated here. Let's actually remove them and just see with our simple product, I see some errors here. Some problems with faces. Okay, let's use graphic and now let's try, maybe not digital art, but something else. Let's do a cartoon. Okay, for color, in tone. Maybe choose vibrant colors. Okay, let's generate. I like the color choice here. The brand colors work really well here. Again, we have problems with facial features. Let's actually, instead of cartoon, let's try the three D art. I'm curious what this will bring. Okay, let's refresh because we're again working with the same images. Okay, I think this is a bit better. Again, something wrong with the nose here when it's definitely has poor rendering of facial features. Another thing I want to try, let's delete the three D or here. There's some materials. There's layered paper, clay, and origami. The images I saw from the gallery, the origami was really nice and layered paper. Let's try layered paper again. We have the vibrant colors and I'll have to click it twice, so we have a different set of images. Let's generate and then click again. Let's click again. This layered paper creates a very interesting style. I actually like this style, but I would have to improve the facial features here. Now let's move on to Anime and see Firefly can produce good images with that style. Let's try that. And here we have graphic for as creation. We can change maybe to portrait. Okay, let's see, we have layered paper chosen. The lad paper reflects the style. Now, instead of layered paper, these images are pretty good actually. Let's change it to cyber punk. Let's remove the laid paper. You can actually combine the styles if you want. For example, if you want to choose more than one style, you can. You can choose laid paper, for example, yarn, metal, and so on. It's going to give you a combination of those different styles, which is pretty cool. Okay. I don't want that laid paper. Let's keep the cyber punk and vibrant colors. Okay, so actually I'll save this image. I think this one is pretty good. Okay, let's generate overall the images I'm getting for this prompt are okay, nothing too spectacular. I would still suggest, if you want to create anime images to use mid journey because it has that specific style. Or you can use the fine tubed stable diffusion model for anime. That would work very well too. But here it's definitely needs some work. Okay, let's move on landscape. Here I have digital art of magnificent medieval castle style. Let's make it fantasy. I'll remove the cyber punk, okay. Maybe not vibrant colors. Let's make the composition, let's make it wide angle for colors, just keep the default color. Okay, let's generate that. Okay, for aspect ratio, let's do wide screen, okay for landscape. We've actually got some stunning images here. Here is a very nice illustration of the castle. The other one here, we have cropped image the same as here, but these two look really good. I like the lighting and just this illustration style that you can use for a fantasy book for landscapes firefly did actually a really good job here. Now let's move on to our final prompt, that concept art. We'll challenge Fire Flight to make some artwork. Here I have the meaning of life, breathtaking, standing, high resolution, highly detailed, inspirational. I'll just keep it as is the only thing. I will remove all those styles for now. I'll remove Fantasy white angle. I've cleared all the styles, now here I want to try using psychedelic style. And here I want the default color, default lighting in default composition for the aspect ratio. Let's choose the default on E the square. Okay? Okay, very interesting. I like the style and the color. So here we have blue and pink and Rg between them, like Yin and Yen. Okay, the interesting fact about these images that all of them have some a circle or a sphere, maybe like a circle of light here, tree the moon and so on. Some of the details are not sharp, it's hard to know what this is. It's probably an artifact, but just the overall composition is pretty cool. Yeah, especially the last one. I don't know what this says, but maybe that's like a tree that makes up a human figure, I'm not sure, but looks cool. Definitely the psychedelic style is interesting. Let's actually try a few more or we can even combine psychedelic with science fiction and let's see what will happen. I like these images way more than the first batch here. Especially like here, it looks like maybe a mirror with the road somewhere. Then the moons. Some surreal art. Yeah, definitely surreal. Maybe some ally here with the eyes and this figure. Here we go. We've tried different prompts. We've tested out the firefly model and we found out that firefly is not best with portraits or facial features. When we did the, the photograph, the realistic photograph, the features were distorted. And then when we tried with the girl riding a bike, that also had some problems if you're using some characters firefly right now as is may not be the best tool, but for things like logos and illustrations, it might be the tool for you see for yourself and try it out. 66. Generative Fill - Logo Editing: Let's move on to our next tool, generative Fill. Here, it can remove objects and paint in new ones from tech description. Let's try it out. Okay, we can try some examples where we can upload our image. Let's upload our image first. Let's actually start with the logo. Here on the left panel, we have Insert and we have removed. Depending if you want to clear the area and insert something different. With your prompt, then you would use the insert. If you just want to remove a specific element or text, then let's choose Remove. Okay, here, I want to remove this text here. Now I can clear this up and then just click remove. Okay. Here on the first image it kept created a different text, I'm not sure why, but on the other ones, it nicely removed the text. On the second and third, the same thing here. On the fourth one, it added some text. Maybe it thought that it should have the text here. Let's keep this one. Another thing is if you want to insert a different text, then we can try to add it here. Let's say now again in this area, I want some text. What I'll do is, again, choose the area where I want the text to be. Here you can describe the image that you want to create. You can only use English for N L. Here I want the text. I'll put text. Okay, let's generate. As you can see it tried to put some of the letters but not quite. In all of these, we see that there are certain problems. The thing here, let's cancel this. We have this settings, we can adjust settings to help us with this prompt. The first one is mat shape, and that's basically matches the image to your selections shape. Basically, this is my selection shape. If I want the image to be exactly as my selection shape, then I would conform. If I'm more flexible where the image would be, then I can put the free form. In most cases, the free form would be more useful because with this tool, it's harder to make very clear borders. The free form would work better here. The next setting here is preserve content. That allows you to choose how much of original content will be kept in the generated image. If you want to keep some of the original image in the selected area, then you can drag this slide bar to the left. If you don't want to see any of the original image, then you should have this slide bar to the right. It will have only the new content. I think an important one here as well is guidance strength that determines how closely the generated content keeps to the prompt. If you want the generated image to be more closely to the original image, then you can keep this slide bar to the left. However, if whatever you write in your prompt, you want the image to reflect more of your prompt, then you should move the slide bar to the right. Usually it's nice. I mean, it really depends on what you're doing, but I like to have it in the middle or more to the pro to the right here. Okay. In this case we don't want any of the original image. We want it to be as close to the prompt as possible. I will move that to the right here. That's the maximum here. Now let's generate and see if there will be any differences. The images we've got here. I think these are a bit better. It's more closely to our text bakery. However, if we look at a few others, still it makes mistakes. I would say that this model is still not advanced in determining and generating text images. Okay, for now, let's not use it, let's cancel it and just clear. Okay. Now let's try something different. We have our logo that we've generated with Firefly here. We can improve it a little bit here. I don't like those lines in the tree. I want to make it more simple. I will erase it. I'll raise the part I've raised, the area that I don't like here. I will write logo geometric, simple shape tree. Okay, again, let's change some settings here. We don't want to have an original content, so I'll have a new content. And then the guidance strength, I want to have it aligned with the prompt. Let's make it maybe somewhere in the middle and see if it will work. If it doesn't, then we will make it even more towards the prompt. Okay, let's generate that. Okay, added some colors here. Let's try a few more images. Well, this one is interesting. Let's see other ones. This looks also pretty good, but it has a few lines right now, I will actually make it a little bit more clean because we have a few things left. I will remove these things to give it more space for creativity. I'll remove these ones. I'll just keep the branches and remove all of the top of the tree. Okay, here I will also add simple color or two color. We have a basic now let's move the guidance towards the probed here. Let's generate. Okay, let's see others. This one is pretty simple. This looks geometric. Actually, This one is not that bad. I think we have the winner here. I like this one the most. Maybe I'm not the fan of the colors, but the shapes look really good here. Let's check the other ones. This is not bad as well. Okay. But my favorite one is this one. Let's keep it, then I will show you how we can change the color scheme of this logo. Okay. A cool thing here is that you can also replace the background. It's pretty simple. You just click on this background button and it's going to remove the background automatically. It will cut out your object here. Now we can create a different background. For example, back white plan, one color background here. We want it more closely to the prompt. We want it to be aligned with the prompt. Let's move it to the maximum here, let's generate. Okay, we did get plain one color background. It's white, but that's okay. As long as it's plain, it would be easy to remove. Now we have some yellowish, grayish and dark blue. I think this one looks the best here. So I will keep this one. Okay. Now you can download your image and as you can see here, it's going to have this water mark as well. Okay, now what we can do is actually I have another logo that I want to edit and that was from Mid Journey. I think you remember the logo. We have some problems here. I want to see if we can improve them with Firefly. First of all, I don't like the tree trunk tree roots here. I will remove them. I think he remove choice is better. So I'll just select those things. I want them removed here. Let's click Remove. And we've got four different selections how we want it to be removed. I think the best one is number two, but let's see if there are any other options. Yeah, I think this is the best one. Now, let's work on, let's keep this, let's work on this area here. Here, I think I will insert. I will delete all of these things here. I made a mistake here. If you make a mistake, you can always click on the subtract and choose the area that you do not want to remove. I do want to remove this area now with this tool, I corrected my mistake here. Let's go back to the add and select more of this area. Now let's write our prompt. We'll just keep it simple tree logo. Let's see the settings. We don't want any content, we want the guidance strength to be yeah, somewhat close to the original image though it repeats the pattern of the original image. Let's try that. Okay, we've got some extension of the tree. I don't think it matches quite well but it's really try to stick to the same style. I think it adds a lot of details. I'll just use the remove button. I'll cancel, cancel, and I will move to remove here. I will remove, as you can see it now, whatever it proposes, it gives me better results than the Insert button. Okay, I think this one is quite simple here. Yeah, I like this 11 problem is the things, but it's easy to remove. I'll keep it. I will remove this area. Let's remove it here. It nicely cleanly removed it. I'll keep this image now. As you can see, it's getting better here. I think I want to change the shape. Maybe here I will try to insert tree logo. We've got this bird here that actually looks quite fun. Let's actually keep the bird. I'll keep it here, but I'll this part here. I'll choose remove, quick remove. Let's keep this one. Okay, now we've got our dream. Let's work a little bit on this reflection here. Now we have nice outline here. Okay, the last thing to do here is let's keep this. I want the background to be different. I'll again click on the background. If you remove the background, it's just give you some random background ideas. Let's just see what it comes up with. As you can see, it generated quite random backgrounds, but I don't want that. Instead of remove, I'll go to Insert here. I want to specify the background that I want. I want one color plain background. Let's generate that. Okay, didn't quite understood me. Let's go back and one color plain background here in the settings. Let's move it to the prompt. One color orange, plain background. Okay? It actually extended our logo a little bit here. I think this is the best one. Okay, let's keep that in the same firefly. I'll show you how we can change the color scheme. 67. Generative Fill - Portrait & Product Photo: Okay, now let's go back and work on some other images. Here are some sample images, and let's try to work with this lady here again. You can remove the background. Another thing you can do is you can invert, you can delete the subject or the image and keep the background. Let's say if you want a different person, you can put maybe a guy in a blue jacket. And then let's generate that. We've changed the person in this image. I don't think it did a good job here because we have all these artifacts that it tried to incorporate the outline from the previous image. It added some strange wires and the background effect. Not sure what this is. Okay. I don't want to keep that. Let's let's go back here. Not only you can change the subject, the background, but you can also modify some elements. For example, I don't like this orange jacket. I can go ahead and change that. Instead of the orange jacket, I want a stylish yellow jacket for the settings. Let's put a little bit towards the prompt. Let's generate. Okay, this is quite nice. That looks quite natural here. Let's see other images. Yeah, not bad at all. Has even some logo and hot cuts. I like the first one more. Okay, let's keep that. Now what I want to do is I want to change the background instead of this cafe background. Let's remove it and put her in a completely different setting on a mountain top background. I want it with sunset view. You can see that the background, it doesn't quite match with the person here. The big issue is the lighting. The lighting does feel wrong here. If you were to post this image, I would say this is clearly a Photoshopped image. The lighting has a big effect. However, if we cancel and try a different background, maybe let's, let's put rainy city, street background. I want it to be blurry. Let's put blurry, blurry. Again, with these backgrounds, it doesn't feel natural. Let's try a different image now instead of the person. Let's try our product. Let's go back and let's plod the image of our sneakers and see how here we can replace the background. I've uploaded the PNG image, it doesn't have the background, but now we can just insert the background. Let's click the background, and let's put sneakers placed on the top of the snowy mountain. And let's generate. As you can see here, it was nicely placed on the grass or the rock with snow. I think number three is the best here. Let's try a few others. Let's cancel and instead of the Snowy Mountain, let's put some fun background, maybe color splash, studio background. Okay, actually here we've got nice shades. It integrated the product with the background or this one is cool. I like the splash. Let's try a few more and see if it comes up with more interesting results. This one is very contemporary splash here. I like this one. Let's keep this one. I will say that here again, we see this water mark, but the quality is pretty good. It has a high resolution compared to even the creator kit. As you can see, this generative fill is a very useful tool. You can use it for logos, for portraits, remove, replace certain object elements that you'd like or you want to improve. Or you can replace backgrounds for products that would work here very well as well. There's so many things that you can do with general. Clearly, it's free. Why not test it out? 68. Text Effects: Now let's move to our next to text effects. This is the tool unique to Firefly. I didn't see anything like this on other platforms. It generates really cool typography in the gallery. Here are some works generated with this tool. Here we have fur wires, popcorn that we have, moss, gold dripping paves can make those cool effects. Okay, let's try it out here. You can enter the text, what you want to generate. Let's say if you want a letter H, then you can put H. Or if you want to generate, I don't know, high, then you can put high here. Then describe the effect you want to generate, for example, butterflies and flowers. Let's generate. We bought these two letters and I. Here are four different examples. You can choose between different variants here and select the one that you like the most. I think this one is very nice here. Then in the right window, you can choose the font that you like. Right now it uses a Man Pro, but there are other fonts that you can choose from. I don't see anywhere to applaud your own font, but here you can choose between these fonts. And if you're using Chinese characters, here are here. Okay. Now you can also choose the background color. Right now there is no background color. If we download this, let's see, it's going to be a PNG image, it wouldn't have the background, but let's say you want the white background, then you can put the white background. The text color. In my experience, it doesn't do quite much. If you choose, let's say the green color, it would change it a little bit, but here I don't see much of the color. Anyway, it doesn't influence too much because the texture and colors comes mostly from your prot we can keep it default just for curiosity, choose maybe a Chinese character. Now let's use a Chinese character. As you can see, it knows the Chinese characters and it filled with our prompt here, which is really nice, Like the butterflies here. Okay, let's try something different now. Let's try a few more text effects. Let's try, maybe underwater. Let's click January. This is quite fun. So we can see some fishes here. Corals that looks interesting. For example, if you are making illustration for a book and you want the first letter to be a drop cap, let me show you what I mean. Sometimes the books would have this illuminated letter that is a drop cap that has illustrations and ornaments. Then you can use Adobe Firefly to make these letters. As you can see, it makes really cool effects here. Okay, let's try a few more. If you don't write your own text, by default it's going to be firefly now instead of underwater. Another cool one is the jungle. Here you can choose from sample prompt. And the sample prompt actually has that one. I really like that one, the jungle ne. Here we have some jungle Ne inverts. Okay. Another one I want to try with more magical one. If you're maybe creating a fantasy book, then maybe we can use that one. Let's try fairies in a magical forest. Okay, we've got very interesting illustrations here. We've got some mushrooms, a fantasy forest. I don't quite see any, but the design looks very interesting here for the last prompt. Let's try. Here we have the box. I would say that more abstract textures like maybe flowers, leaves, what else is here? Lava wires would work better with the text, because right now with the box, you actually can see a lot of artifacts that don't look too good. That is the tool that allows you to do text effects. 69. Generative Recolor: The next tool is generative recolor. Here you would need to have an SVG image. You cannot upload P or PNG, it has to be SVG. But for that you can always use the vectorizivelready covered that. I'm not going to spend more time here, but what I did, I've changed some of the logos that we've created with Firefly and in Journey, I use the vectorizi to convert them to SVG vectors. Right now I can upload the SVG files here I, let's say this one was the first one I've converted to SVG. Now I can describe the color palette that I want. This is the logo of a tree in a water droplet. Let's change it to blue. I want it to be in blue and green color palette. Let's generate here. It gave me some ideas here. On the right hand side, we've also have sample prompts. Let's see some of ideas here. Dark blue, mid color palette, yellow submarine. Then we have Terracotta desert. Then we can choose the color scheme in harmony. There is default complementary colors, analogous colors, triad split, complementary, and square. For mine, I want colors to be very similar to each other. I will choose analogous here. I can actually choose what colors, blue, purple, and light purple. Okay, Now the image starts to look more on what I want here. I think the number two looks really as. Actually I'm going to say that one. Let's just see what other Sambal products that we can use. We're using the tract desert. Let's see, faded Emerald City, lavender, storm, summer by the sea. Well, right now they all look quite similar. And the driving factor here is the analogous color scheme. Along with the colors that we chose. Even though we choosing those different effects here, the color scheme does look quite similar. Okay, let's now move into a different vector image. Let's try a different logo that we've used. Now let's upload our logo that we did with Mid Journey, but then we've changed and edited with Firefly. This is our SVG file here. I wanted to be in blue turquoise palette. Okay, let's generate this. Okay, now you see that we've got this turquoise color prevalent in the image. Okay, The only thing I don't like here is that I want this drop to be white. I can change that by choosing the color. Let's choose the white color here. And now we're getting what I want. We have this water drop filled with white. Now, this tree looks really nice on this white background here. This was the blue turquoise, let's say if you've changed your mind and you want to experiment with other prompts, then we can check out and see maybe some other color scheme would be nice here. Let's try salmon sushi, for example. Now we can see that these colors are also pretty cool here together. Let's try the Turquia Desert. These are more pastel colors. Again, blue and greenish color scheme. Lavender store, we have this nice purplish color scheme here. Then we have the summer bite, the sea. All of the colors I think, look really nice together here. And let's say if you are struggling to find the color scheme, then you can use the stool to help you choose the colors that will look good together. Now let's go back here. We have other sample images. For example, you have an artwork, not a logo, then you can also recolor the whole. Here, for example, we have this image here. We can try it out here. It used the soft pastels. Prom's. See if we experiment and change the color scheme. Right now it's default. Let's choose complementary colors. Now we have the complementary colors, red and green, orange and blue, and so on. Let's change it to analogous. For analogous, we can see that the colors are more similar to each other. Let's move to triad. Now here we've got the triadic color combinations. Let's see split complementary. Then we have a square color scheme. Here we can try to refresh and every time it will be different colors depending on what color scheme you want. You can choose that particular color scheme. You can write your prompt and choose the exact colors that you want to be part of the image. That's what generative recolor will do for you. Instead of you going and manually changing each color, It helps you to recolor everything in a totally different color scheme. And let you experiment with different colors, combining different colors together, and so on. So this is it for Adobe Firefly. In this model we've covered a lot. We've talked about text to image generated with fireflies model and we've used different prompts and we've generated interesting images here. Then we've talked about generative fill and we found out that not only you can do in painting, but you can also remove certain objects or elements from the image and also replace the background. It would also work for product photos where you want to change the background. Then we also tried to generate cool text effects. This is a fun tool that allows you to create the illuminated letters. Then we went on to talk about generative recolor, that when you upload your SVG vector image, you can make it in any color scheme that you want and try out different color palettes with that image. For now, these are the only tools that are available on Adobe Firefly. And we can see that there are more tools that are still in development. This is it for this module. In the next module, we will cover another very interesting platform called Runway ML. See you soon. 70. RunwayML Introduction: Hello, in this module we will be covering another very exciting platform, Runway. Runway is an AA video and image editing app. It was founded in 2018, and initially it was launched as Models directory that allowed users to deploy and run machine learning models. Currently, it's a platform that offers multiple AA tools like AA image generated in painting and other image editing tools, but its main focus is on video editing models. Runway was also involved in the development of open source stable diffusion. Along with stability AI, and a few other universities and companies. I actually wanted to find what model do they use right now for AI image generation on their own platform? I couldn't find this information anywhere, but it is likely to be a proprietary model based on stable diffusion. Right now, Runway has also Runway research that partners with universities to research new AI models and publish papers. Then it integrates findings into new products. So to sum up, Runway plays an important role in AA research, especially in areas such as filmmaking and video editing. In this course, we're not covering video editing, but we're going to check out what Runway has to offer us for image editing. Here are the tools that Runway offers for image generation and image editing. The first one is the text to image generator. Let's go to Runway to check it out. Once you sign up, we'll go again to Runway. Under images, you will see image generating tools and image editing tools. Right now we're in this generate images section and the first one is the basic text to image generator that uses their own proprietary model. Then we have image to image generation. The third tool, we haven't seen that in any of our platforms that we've covered so far. This tool allows us to train our own models. It allows to create a custom model, portraits, animals, styles, and more. We will talk about the tool more in future videos. Then we have infinite image. That's basically out painting. So you write your prompt how you want to extend your image. For image editing, we have expand, expand image Is quite similar to this infinite image. What's the difference is that with the expanded image, you actually need to provide the prompt. But with expand image, it automatically expands the image. Then we have frame interpolation. Frame interpolation is a tool that turns a sequence of images into an animated video. If you have a bunch of images, you can upload them and make a video on it. I will also show you some cool ideas, how you can use this tool. Then we have a race and replace. That's basically the same as in painting. Then we have the backdrop mix. This is the same as background replace. We upload the image, it removes the background, and then we can choose a different background, we can generate a new background for red. Then we have image variation. We upload our image and then it creates variations of that image. The next tool is actually very cool, it's add color. If you have black and white photos or images, you can upload them here and it will actually colorize the images here. I'll show you some examples how to use this tool as well. Then the last one here is the upscale image, basically that upscale the image. Okay, then under more there's actually three D. And I wanted to show you this tool. So it's just one tool and it creates a three D texture. From prompt, we'll try that as well in the next video. Rightway does have quite a few tools. We'll check them all out, test them, and also compare some tools to the tools that we've already covered from other platforms. 71. Text to Image Generator: To use runway, the first thing you'll need to do is to go to the website Runway Ml.com and sign up. Once you sign up, you can log in to the platform. Here you'll see a lot of different tools. Here is your profile. Here are AA tools. Here we can see popular AA magic tools. This includes the video editing and image editing. Here, you can also check out some tutorials that were made by Runway on their tools. I also want to mention that here I have the paid plan. Some tools are limited if you don't have a paid plan. If I go to my account here, manage your plan here, you can upgrade your plan. When you sign up, you'll get some free credits, but they run out pretty quickly. So if you want to upgrade, you can buy here and you can update your plan. Here are some plans. You have the free plan, standard and the pro plan. Right now I have the standard plan. Okay? Now let's move on to image generating tools. Here if we go to images and then here we see generate images. The first one is the image generator, let's try that one. Based on my experience, I found that runways image generator is inferior to other platforms that we've already tried. I'm not going to spend too much time here, especially that I think it's quite pricey to generate images with the plan compared to other platforms where it's way, way cheaper per generation. It's quite pricey here. Especially here, I paid $15 and I only got 200 generations, not that much. We even covered the clip drop that gives you 1,500 generations, Way more than this. So keep that in mind. Okay, here, let's just try our first prompt and you'll see what I mean, that it's not that great. Okay, so we have our professional portrait, and then here we can choose the ratio square, white screen, landscape portrait, then resolution. If you're in the free plan, you can only choose the 512. If you have the paid plan, you can choose a higher resolution. Let's, let's choose the regular thousand 80. Then you can choose how many images you want. Let's choose four. Then here we can go to Advanced Settings and choose Style, Futuristic, and so on. So that's the style. Then we can choose the medium here. Canvas, airbrush, graffiti, drawing here, or oil painting here. I will choose photography, then moved, let's put Beautiful. Then we have the prompt weight to remind you. Prompt weight affects how much prompt do you want to be reflected in the image? It determines how much to take the prompt into account in the generation. Higher values may result in more precise results. Lower values will generate more creative outputs. The standard prompt weight value is around seven. Here we have 7.5 that's standard one. Let's actually move it maybe eight, a little bit more here. Okay, it's more aligned with our prompt here. Then we have our set right now. We're not changing the set here. Okay, and that's it. Let's click generator. As you can see here, we've got some poorly branded images here, we've got a lot of artifacts, especially with eyes here again, the portions are distorted. Even though we have this nice and long prompt, we're still getting not great results. This is for portraits, and you'll find that using faces with the current model right now is just not going to give you great results. Let's try something else really quick. Let's try maybe a logo and see if that's going to work. Here we have line logo. Let's change the style. Right now, I will choose digital or minimalism. Let's choose Minimalism Sum, let's choose illustration for mood. Let's keep it to non default one prompt, let's make it higher because this is logo, we don't want it to be too creative. We want to be exactly as our prompt. Let's generate. Okay, we're getting some simple images here. This one is just a circle with a cherry cupcake. Another cupcake and just the cherry. Yet, not my favorite model to work with. But again, let's just do landscape. I think landscape should be fine here. Landscape usually works with any model because it doesn't need precision. It can tolerate lots of mistakes. Let's use digital art of our medieval castle. Then here again, we have four images as output. Let's pump our resolution to two K. Then for the style, let's fantasy, for medium, let's make it into oil painting. Here's our oil painting and mood. Does Epic. Oh yeah, it does have Epic. Let's use Epic. Okay, prompt weight. We can now make it smaller, maybe six. It can make some creative images here. Okay, let's generate it here. Let's check out those images. It's okay, it's cropped here, that's not good. But other ones, we're getting shapes here that doesn't look like castle here. Okay. So it's a little bit messed up still. Here we have something flying in the air. This is definitely not advanced model compared to other platforms. Okay. But it does have cool tools. Let's check out other tools. 72. Train your Own Generator: So let's actually cover the train your own generator first because this is a pretty fun tool to use. Okay, here you can train your own model. So you'll basically need to upload similar images of the same person, animal, object or style and that basically teaches AI about that person, animal, or object. So the next time you generate images, you can actually generate images with your own face or with your pet, or with some with a certain object that you want. That's how it's used here. You can choose from portrait generator, animal generator or custom generator. Let's say if you want to generate images of yourself, then you can use train a portrait generator here. Just click here here, it's a paid feature. But if you have a paid plan, you have one free training. We can say train portrait generator. Okay, here you will need to upload images of yourself and it should be 15 to 30 images of a face with different backgrounds. You can't redo this later, so choose with care. This is very important to make sure that the images you choose are high quality. That will help AI learn your face better and the images you'll get will be better. Here you can check out some image examples. Here are some selfies and as you can see, different backgrounds. I would not recommend using cropped images because that will affect the output images. This is the input. If we check the output, you'll see that some images will be cropped and that possibly because the input has cropped images. These are the images that you give AI for training. So make sure that they're good. Okay. You would also want to make sure that images are cropped in square one to one for best results. Of course, avoid inappropriate images. Okay, let's upload some images, for example this one. This one. For this image, it's not a square, I would have to make it into a square. And then upload it. Right now these two are squares and then we'll need to make that square as well. Okay, here with Mac, I can use Preview for example, but you can use any other tool here. I can choose a square, let's make sure that it matches. Now we have a square. Can move it a little bit, now Can crop okay, and safe. Now it's a square and I can upload it on, you'll find 15 images. You'll upload them here. And then you'll click Generate after half an hour. And so you'll have a new model. Let's get into the model that I was able to generate with the images. Okay. After waiting some time, I was trained on my images and here are the images that it generated. We've got different styles, for example, black and white pastel colors, and so on. Illustration fantasy, my only concern is that it didn't capture my eye color or even hair color. There's a lot of black hair here, but it captured my nose here. Overall, I think that it did not a bad job in training and generating images that look somewhat like me. But honestly, there are other tools that, in my opinion, train models way better and cheaper. But we'll also talk about that in the next few modules. So make sure you check those modules out too. Okay, now we have these images. We can actually make more images. Let's say you like this style and you want to generate more images in this style, you can. Okay, for that you need to go back and you need to go to generate images. Here, you can choose the text image generator. Here. You just need to write your prompt, what you want to see. Let's say pop art and then I'll describe myself. So young woman with curly blond hair and I want to get colorful. So I'll just put colorful here. You can choose default or you can choose something else here. If you have trained a model, you'll see the name of the model. For some reason I've decided to name my model. I'll just choose K, K and see how it added here. Now it would know that it needs to generate images with my face on them. Okay, let's click Generate. Here I have low resolution, let's change that. I'll change the resolution to, let's put two K, the number of outputs, Four then here. Okay, you see it used, in a way, my facial features to create this image. Okay, let's then change the advanced settings for style. Let's choose Pop Art here, and then Mood. Let's make it beautiful. Okay, and the prompt weight. Let's keep the default one. And let's click Generate. Again, as you can see it, try to integrate my facial features here even though it's not quite successful. But again, in the future module, I will show you a platform where you can also train and generate images of yourself, an animal, or of objects or styles. It is my opinion that platform is more successful then what we have here, but let's try one more time and let's move on to other tools instead of port pop art. Let's actually here we have comic art here in the prompt. I'll just add also comic book. I'll leave everything here and I keep my model chosen here. Then medium for mood maybe. Let's choose colorful. Let's generate here. We've got some comic books here. We have a lot of images on this one. Let's expand again. The facial features are totally wrong. Let's see other ones. This one looks like a human, but still the proportions are bad. This one is, I think, the best out of all of the images that we've generated. But again, still lot of artifacts here. Let's say we'll save it in downloads. This is the image and that's supposed to be a two K image. Okay, let's go back here now. We've tried using text to image generated with our own model. The results weren't great, but now you know the concept and I'm going to show you other platforms where you can also train your model and generate images based on your model. 73. Image to Image and Infinite Image: Now let's move on to our next tool, which is image to image generator. And let's try it out. Here we have our ballerina. Okay, here we have the image. Let's put our prompt. Let's ballerina dancing in a magical forest. And then I want trees and flowers in the background. Let's put it beautiful. Okay, so here again, I want four outputs. Resolution. You have the H, D, or D, so let's try it out. Okay, here it got the faces terribly wrong and we have an extra limp here. There's actually no way you can write the negative prompt. Okay, then we have number three here. The third one is not that bad, even though the face is really bad. But the proportions, and the proportion is pretty good here. Here again, we have the same bid posture as the original image, but the face is horrendous. Let's actually try it with a different image. Now that we know that it's really bad with faces, let's not use any more faces here. I have a wolf here. I'll just put iridescent wolf magical colors. Again, the resolution is, let's try that. Okay, here we actually got the iridescent effect on the wolf. The wolves proportions are pretty good here compared to human faces. Let's see other ones. Yeah, it captured the fur eyes nose. Then we have the jaw here. This one is the best out of all the images here. This one is pretty good. Okay, another thing we can do with image to image generator, we can actually use our own model. Let's try Apple for example. Here I have a art. I wonder if I can put my face on it. Here we choose this image. Okay, now I'll just put myself a young woman. Let's add the pop art art. A young woman with curly hair. Now let's choose our model. Let's generate. Okay, as you can see, it used the original image and it also captured facial features like a nose, maybe my eyebrows here. Not too bad. Not too bad. This is something that you can do with image to image generator. If you have trained your own model, you can try out those things. Now let's move to our next tool, Infinite Image. Here you can generate a new image. Basically just write your prompt and click Generate, or you can upload your own image. To do that, you would need to click Add Image Pattern, and click Upload from Computer. Here, I prepared some artworks. Let's try this one. This is the great wave of Kanagawa. Okay, here we have our frame. We can move it anywhere where we want to extend the image. For example, I want it to be here. Okay? Now all I need to do is make sure that it overlaps with the original image. Then let's put a wave in the style of wood block print. Okay, this is the wood block print style. Let's try that here We've got different variations. This is the first, second, third, and then I think the best one would be the second. I think the second one is best here. Because now we have this. Can Sky, let's accept it in a similar way, we can move it here. And just change our prompt. A wave here. Let's just put a sky in the style of wood block pred. Here you can see that it added some characters which probably don't make any sense at all. Let's actually the reason is because in the previous generation that we had, we had, let's cancel this. We had some characters. It took the idea from here, but we don't want this because I generates some nonsense. Let's use as we can now erase this part. Okay, let's make the race a little bit bigger. Okay, I think this is a good size, so just remove it now. Hopefully it's not going to generate more characters. Okay, let's try this. Okay, now we have something different here. Maybe I'll go with this one. Okay, here we've got some characters. Again, let's use this one, Okay? Now you can also add more images if you want similar as Ali, you can add multiple images. However, the difference is you cannot move the original image around. In order to upload the image in the correct position, you have to move this frame in the place where you want your image to start. If you want to start the image here, then you need to move this frame here. If you want the image to start here, then you need to move this frame here. Okay, let's upload our other image. Let's put it maybe here. Okay? And then again, add image from computer. Here I have the artwork by Salvador Dali. Okay, now we have this image here. Again, if you change the mind and you don't want the image to be here, unfortunately you cannot move it around. You would have to go back to coma or control Z and then upload it. Move the frame again and upload the image again. Okay, so that's a bit of inconvenience now, let's try to merge those two images together. Let's erase some parts of it, for example. Let's erase those sharp edges here. Okay, now maybe here as well. Here. Okay, now let's move our frame somewhere where it overlaps with the two images here. Let's try this. Let's put a way that becomes a scarf, and then here I'll put surrealism. Okay, let's change the setting, maybe for prompt weight. Let's keep that somewhere in this default one. Okay, let's generate. Okay, let's see here. I think it merged pretty well. Here we are coming from this woodblock print to surrealism, and here are some variations. And it also added the C in the background here. Okay, maybe the first one was pretty good. Yeah, I think the first one is the best here. Let's accept that and I'll quickly do some more. Okay, here is the final result of merging those two images together. I added a few frames here, and in the bottom here, I think overall, it's pretty good here. If you want to save the image, then you can head down to this button and click download. Okay, we're done again, if you want to undo or redo, you can use these buttons or you can use the keyword control or command team. 74. Image Expansion: Okay, let's check out some other tools. Let's go back to Generate images here. We've tried all the tools here. Now let's move on to Edit Images. And the first one is expand images. That's very similar to the infinite images that we've already done, but with a slight difference. Okay, let's check out the difference here. We just need to upload the image. Let's use the same image. Let's use the wave. Okay, here we have our image on the right. We have settings here, we can choose scale. It's basically how much we want to zoom out. Right now it's one x, that's the original image. If we want to zoom out a little bit, we can choose 0.75 x, and even further is 0.5 x. Let's keep it maybe. Let's first choose 0.75 x for aspect ratio. We can also change the Ascra ratio, but let's keep the original one. Okay, You can also write the prompt if you want here. It actually automatically generated the prompt based on our image and right now it's correct. So it's a Japanese painting, so maybe it's sort of painting but woodblock print showing the great wave of Kanagawa, Japan. As you can automatically does the prompt, so you don't need to write it yourself. You can if you want. Okay, let's that. Okay, here are our zoomed out image. As you can see it in all of these images, it added some characters which we do not want. If you don't see certain things like characters, then we should have prepared this image beforehand and removed the characters. Okay, And there is actually a way to do that here in Runway. So if we go back here, it images, there is a race and replace. We can just add our image here. Now we can use this eraser to remove the part that we don't want. Then I'll just put a sky wood block print and then maybe Japan, Japanese style. Okay, let's generate that. Okay, here it added some more characters. Okay, I don't want to see any characters actually now because it takes the information from the original image. I'll move my prompt weight to maybe 26 or 20. Let's move it too high, maybe 20. It's more aligned with our prompt. So it's like maybe I'll put simple Simple Sky. And I'll use more of this area around. Okay, Simple yellow sky woodblock print Japanese style. Let's remove Japanese style so it doesn't add Japanese or Chinese characters. Okay, let's try that. Okay, let's see. Okay, the second one is a bit better, still not something that I really want. For this case, I might go and use the clip drop to remove because here there is no removing clean up tool that just nicely removes the thing like it must add something to it. Actually, I'll just go to clip drop here. In the clean up, I will upload my image here. I don't want to see these things. Let's clean. Okay. Beautiful. That's all I need. Okay, let's download it, save it and go back. This one, this is the runaway. This is okay but see it added some yellowish pinks that don't work with the style. The clip drop version was way better and faster. Let's use that for the expand image. Now I can choose the clean image. Okay, here I don't have any characters. Hopefully it when we zoom out, it's not going to add those characters here. Okay, let's use the scale 0.5 x prompt painting, that's showing waves in the ocean here instead of painting. Output block print. Okay, let's generate. Okay. As you can see right now, it didn't add any more characters, so we don't have any more mass. Well, added some text here, but for other ones we don't have it. Which is way nicer. This is something that you need to take into account. For example, let's expand this image. And this one is pretty good. We can download this. Okay, let's also try this expand tool with our, the image. So we have our am just let's try it out here. I get a 10.5 X for prompt. It automatically generated a man standing in front of a woman with an angry look. That's not an angry look, but let's remove that. A man standing with a woman beside him, looking back, had another woman. Okay, that's a better description here. Let's generate. Okay. Runway doesn't like my content. It triggered some moderation guidelines. Okay, I don't know, what did I say, but let's remove that and just not use it. Let's see. After a few attempts, it still didn't work. However, a day before when I tried it out myself, it worked perfectly fine. So this is the image that I was able to generate with runway and I chose at that time the best image. And here is the image that we've got with the crop, So you can see some differences here. In my opinion, clip drop did a better job in terms of extending the image. Here we can see a lot of artifacts and especially if it generates more figures, then there is a big problem with faces. That was for runway. Let's just try a different one. Since we cannot do other meme here, let's use the baby. Let's see those extended images. Okay. This one is not bad. The child is in the sand playing. The problem here is these things. So it should be some kind of toys here, just artifacts here. It's just setting down then not sure what this is and then some huge head here. Okay. The best one is the first one, but still I think it needs improvement. Okay. That was the expanded image tool. 75. Frame Interpolation, Erase and Replace: Now let's move to frame interpolation. This is a fun tool that allows you to make your images into videos. For that, I think we should prepare some images. And the best way to pay images for the frame interpolation, if you want to use the generated images, is the rays and replace tool. Let's first use the rays and replace tool. Okay, here we can use a landscape. Okay, so here I have an image of mountains and the sky. I want to make that the sky move. Let's do that here in the ray and replace. I will erase all of the sky and generate new images with the sky. Okay, let's raise all of this. Can now maybe make it smaller here. I'll put photo realistic sky with clouds. Okay, and let's generate. Okay, so we have some new images here. Okay, well maybe this one matches it a little bit, but I think clouds are too heavy. This style just doesn't match the image. Okay, maybe this one, but still it doesn't feel photos. Okay, let's actually improve this. I'm going to erase those little details. Maybe that will help. Okay, so let's move those little details. Okay, so now for the prompt. So I'm going to have footers steak and then I'll put blue sky with clouds. Okay, let's check our setting. Okay, so our prompt weight is ten, let's move it back to 77.5 Okay, And then maybe some clouds. Okay, Okay, this is way better. Let's see. Other ones, we have some sky here. These clouds are better than the previous batch of images. Okay, these ones are all good. Honestly, I'm going to save all of them. I'll download all the images here. Okay. Then I'll click cancel and I'm going to generate one more batch of images. Okay, I think this one is also very good. Okay. Okay, so let's save them all. Okay, now I'll go back and I'll go to images and I'll go to frame interpolation Here. I will upload all the images that we've just downloaded, these ones as well as my original image. Now I will change the sequenced based on how many clouds they have. The fewer ones will go first, maybe, and then more clouds towards the end. I think this is about it. Okay, so maybe I'll move this image here. The sky is more clear. Okay, so now it goes towards the more cloudy sky. Okay, And then for settings, you can choose the clip duration. The default one is usually 10 seconds. We will try different using different images. Sometimes it's nice to have it less time. Then we have also in the Advanced. How much of transition time do you want right now? Let's leave the setting as and let's generate. Okay, Let's see what we've got here. Okay. As you've seen, we've got the skies moving right now. It feels a little bit unnatural because the transition takes very long time. What we can do is we can change the clip duration. Instead of 10 seconds, let's make it into around 5 seconds. Maybe 5 seconds here. Let's generate. Okay, It feels a bit more natural, but still we see those big traditions. If you have more similar images, then that would work better. Let me show you what I also try doing here. I also used only one image and then I and replace Tool to change the background. Let me show you. I used this image with the rays and replace Tool, I replaced those background mountains. Okay. And I've generated a few images like that one here. We've got some, let me re arrange it again. Okay, Maybe something like this. Okay. Now again, clip duration. I usually like it shorter. Four smaller number of images. Let's make it around five. Yeah, five, then let's generate. Okay, let's see here. You can see those mountains moving. And that is a really cool effect that if you want to do that with your photos. Another way you can use frame interpolation is when you make lots of photos. Let me show you here. I have a bunch of photos. When I add all of them together, now I'll change the clip duration to maybe four, also for advanced. Okay, let's use the transition time, 100% first, and then I'll show you how we can change that. Let's generate, let's see here. Here you can see that transition is pretty long. Let's reduce transition time to maybe 20% okay, 22% I'll also move the clip duration to 1 second. 1 second. Here we go. Let's re generate. Okay, let's see. Now this is a little bit better, and that's how you can combine your images into a video. Okay, another cool thing you can do with frame interpolation is changing the subject of the image. Let me show you. Okay, now I need to go to race and replace. Here I have this image of origami on the plain background here. Again, I'll use the race tool to erase this part here, I'll put origami flour. Okay, Let's generate, okay? And here are some examples. For example, this one is knives, okay? Now, instead of origami flour, let's put origami. I think this one is pretty cute. Let's use this one. You can create different objects using this erase and replace tool. And then if we go back to our frame interpolation here, we can upload all those images. Here we have those two. I also have a few more that I did earlier, so for example, one, now you can change the settings. Clip duration. Let's make it maybe 3 seconds. 3.5 for Advanced. Yeah, let's keep the transition time or okay, let's generate. Let's see, here we have a cool effect where one object becomes the other object with this nice transition. These are some things that you can do with frame interplationase, replace tool. 76. Backdrop Remix, Image Variation and Add: Let's move on to our next tool, Backdrop Remix. That's basically a replace background tool. Let's use it here. We can use our sneakers that we've tried with other platforms. Here are our sneakers. Okay, here for settings, you can choose the scale, so you can zoom out a little bit, or you can zoom in. Then you can choose the style, like apartment bakery, or you can choose a custom one. Then there are for the backdrop for the studio, flowers, beach, and so on. I think for this one, let's try to find some mountains. Maybe outdoors, okay? Mountains here, we can put sneakers placed on top of a mountain rock. Okay, let's generate that. Here we've got some images. I think they're very similar to one another. Let's see. Okay, I think this one is pretty good here. It tried to add something to our shoes, which I don't like. Again, it added some layers. Again, what I found is that when you use runway for background replacement, it may add something to your subject or object. What I tried replacing the background myself with the sneakers Here is the result that I've got. That was for Snowy Mountain. As you can see, it added this extra platform to our shoes that was like almost to every single image. This is compared to Adobe Firefly or other tools like Clip Drop that only add the shadows and integrate the product with its surroundings. And that's it. For some reason Runway adds extra details. Let's try it with something else. For example, a chair. Here, it's a PNG image, it doesn't have background, okay. Here I want it with a zoom out, so 0.5 x. And I want it to be in the apartment. And let's put modern style apartment. And here we have cheer inside a modern style apartment. And then standing beside plans. Okay, let's try that. This is the original image. Look at those legs here. When we go and check the images that were generated with Runaway, you can see that on all the images here, it added extra stuff to our product, which is not good here. It added really long legs and so on. Let's expand it here. Definitely see that plants have artifacts and just the whole setting. The objects are not clear. Here is a little bit better, but again, we have those legs, lots of problems here. I wouldn't use backdrop mix to replace background of a product photo because the images that you will get will not be good quality images. I think that backdrop remix tool needs a lot of improvement before it can be useful. Okay, let's go back. Our next tool is image variation here. Let's try to use the same image at the image that was generated with mid journey and the one we've tried with the clip drop here. For settings, we can only choose number of outputs, 123.4 images. Let's choose the highest, it's going to take the most credits, but let's try it out. Let's generate. As you can see on these images, it's basically the same studio, the same design as the image we've uploaded. But the only differences is just small details. For example, the floor tiling that we have, just the colors of the desk, the chair, and then different monitor. And so on. Slight modification of the image, but overall it's very similar composition. Okay, if we compare that to clip drop, here is our original image with Runway, we were able to generate images that have almost identical composition with slight changes of the design texture and so on. However, with Clip Drop, it actually it kept the same color palette, but it rearranged the object here. Here it has the same style, but now the desk is in front of the window. So these are the differences when you use the clip drop and runway for image variation. That's the image variation with runway. Now let's go to the next to add color. Let's try it out. For this, I've prepared some black and white images, let's check them out. The first one is let's use the landscape here. Okay? Basically we just need to click Color. Here we go. Now we have colorized image. We can see mountains white, and then we have foggy hills, and then some dark green colors. That was the landscape. Now let's try to do the old photo of Marilyn Monroe. Let's use that here. Again, colorize it here. It did a pretty good job. Here we have quite natural skin color. Then we have this blond hair color and red lipstick, and then we even have some silver pearls and this old style sofa. However, here we are getting slight green tone. Maybe that's something that should be different here. Okay, but overall, I think it did a great job here. Let's save it now, let's try a little experiment Here I have a color image of a family in the park. In Preview, I've changed it into black and white image. Here is the black and white image. What I want to do is try the color tool here and see how different it will be to the original color image. Let's see. Okay, we've got some colors, but as you can see, the colors are a bit more pale. We do have the green grass, but some colors are quite different to the original image. And let's check out which colors. Okay, here is the original image. On the original image, we have lots of brightness and saturation of colors on the colorized image. Colors are not saturated here. For example, for genes here it's dark blue. Genes here it's almost violet. As well as with his shirt, almost violet color. And look at the girl's T shirt on the original image, it's bright yellow. Here we've got this pale bluish, maybe a little bit purplish and greenish color. As you can see some of the colors it would get spot on grass, maybe the skin color. But some of the colors like this yellow here, it would get wrong just because there's just a variety of colors that would be the same on the black and white image. Also, saturation may also no match. Here are some limitations with using this tool, but otherwise I think this is a useful tool to use, especially if you have some old photos that you would like to colorize. 77. Upscale Image and 3D Texture: Our last tool is upscale image here. I want to try the same images as we've tried with other platforms. Here, I'm going to have my photo in low resolution. Here is the photo here. I can upscale up to four K, However, you'll have to be on the paid plan to be able to upscale two K or four K. Let's upscale to four K here and let's process it. Let's expand the image and let's um, in okay, it has smoothened out some patchy areas and reduced noise. Overall, it enhances the image. However, I think it smoothened out quite a lot on the face. Now it doesn't look photo realistic. That's why in my opinion, clip drop had the best job done in terms of photo up scaling compared to runway or big GPG runway. In my opinion, it over smoothed the area making the photo less potalistic. However, in clip drop, this wasn't a problem. But I encourage you to try out all of them and see which one you like the most. Okay, then let's try our artwork here. Okay, let's now try, this is the meaning of life artwork that we've made with Mid Journey. Okay, and again, let's use the four K and let's process it. Okay, let's expand and let's zoom in. Okay, Again, a pretty good job in terms of improving the resolution of the image. And now if we look at the moon and clouds the human figure, it doesn't look pixilated. Okay. However, if we compare this to other platforms. So this is the original artwork and then here are our three up scalers, Big GPG, clip, drop and Runway. My favorite one was the Big GPG specifically for artwork upscaling. And that's because here, zoom in here. It added those very nice smooth lines that I think fit very well in this artwork. But again, you may prefer a different platform for your artworks. Now let's move on to our next section, three D texture. Here, under more we can find three D. Here we have the tool that allows us to create three D textures from a text prompt. Let's try it out. Okay, here let's write, for example, mossy texture and let's generate. Okay, we've got this three D texture that can be used for games or other visual elements. On the right hand side, we can change a few settings. The first one is we can increase the resolution to 2048 by 2048. Then for the form, we can change cube to sphere. Also we can have just the image. That's just going to give us the two D image. Let's return it to the cube for tiling it actually at the repeated pattern, we feel one surface. For example, here we just have one image. This is one image, let's say. Then when we increase it, let's say to two. Now this image is repeated four times on one surface. Then when we increase it even further, then it's repeated multiple times. Then when we get to 20, that's the maximum here. Okay, let's make it back to one then. We can also adjust the ambient light if we want to make it lighter surfaces. And we can by increasing this ambient light exposure, the default one is around 40. If you don't want to see that some sides are more darker, then you can change the directional light. We can make it zero. Then all of the sides will be with the same lighting. So if I increase it, so there will be no difference, it would not look like it has any shades. Okay, here you can all the acts which include texture displacement and roughness maps. You can download that after downloading, we'll have those four different PNG images. The displacement, then color and roughness. That's basically all that is to the three D texture. This is it for runway. We've covered all the image generators, text to image, image to image. We've learned how to train your own model. And that we also looked into out painting with infinite image. Then we did some image editing with cool tools that Runway has like expand image frame, interpolation, erase and replace backdrop remix that replaces the background image variation, add color, and upscale image. 78. Leonardo.ai Introduction: Hello, hello. In this module we're going to go back into stable diffusion and actually explore more platforms and more advanced features. In this module, we're going to go and talk about Leonardo I, which is an AA image generator based on stable diffusion. Okay, what is Leonardo? It's an image generator. It was developed to as game artists in creating game assets such as characters, environment items, conceptual artwork, and so on. It was founded in 2022 and the company is based in Australia. Okay, so let's talk about some advantages and limitations of the platform for prose. Leonardo actually gives a good amount of free credits, and what's even better is that those credits are updated daily. So you can try the platform out and see if you like it and really understand how it works. Leonardo has also an image gallery. Let's check it out. This is, once you look again to Leonardo, this is what you'll see here. You'll see a bunch of images created by the community. You can even go to this community feed on the left hand side and you'll see all the images here. Let's say you like any of the images, you can mark them with the heart and they will be placed to your personal feed. To the feed, here are all the images that I liked. So this is quite handy because if you want to reuse some of the prompts, for example, you like this image and you want to use this prompt for example. It actually gives you all the tools to do that. For example, to reuse this prompt or do image to image generation very handy. Also, the images you generate with Leonardo are high quality and you can upscale the images and there's different up scalers which we'll talk about. Leonardo also has many stable diffusion models to choose from, which are fine tuned for a specific style or character. If we go to Leonardo here you'll see featured models and you can see some models here are great for three D animation style for illustration cute animal characters, you can find the model that will be best for your image. It also allows to train your own models. If we go back here in the training and datasets, you can upload your images and train your model. Then it also has an AA canvas where you can do out painting. And in painting, for example, if you generate an image, you can quickly go and edit it in the AA canvas. Also, you can generate images privately with any paid plan, which is great because their started plan is very affordable. Now let's talk about some limitations and disadvantages. Perhaps the first one is the wait list. Right now you have to join the wait list first. After a few days, you'll get an e mail saying that you have the access to the platform. If you don't have Leonardo AI account yet, I urge you to sign up right now. So by the time you want to try it out, you have the access. Another thing is that Leonardo has a lot of features which can be overwhelming for new users. However, throughout the course, we've talked a lot about stable diffusion. We've talked about prompt engineering. We've talked about different parameters. It shouldn't be a problem for you then what I found is that it's hard to produce photorealistic images. With Leonardo, it lacks photorealistic models. Even though on Leonardo we have this model called absolute reality. It says that it's a photostic style model, but when I've tried using it, the images have this game character aesthetic rather than the photography style. All the models that I've tried here, I couldn't quite get that photography style images. The last thing, the models that Leonardo has are stable. Diffusion based detailed prompt is important for better results. That's a brief introduction to Leonardo. And in the next two videos, we'll go and explore this platform. 79. Leonardo.ai Overview: To start with Leonardo. You'll need to go to Leonardo. And if you don't have an account yet, then you'll need to click this, get instant access, put your name and email, and you'll get an E mail. Once you get access to the platform, if you already got an e mail that you're white listed, you can click this. Yes, I'm white listed. And then again, once again, this is something you'll see in the front page. So this is home. These are featured models and the gallery on the left hand panel here we have the community feed. This is the images generated by community. If you like any of the images, you can put a heart here and then it's going to get in your personal feed. Okay, then we have personal feed. These are your generations. Then if you follow certain accounts, you will have the images that were generated by that person. Then the liked feed will be all the images that you liked. Then we have the training in dataset. This is where you can train your own model. Then we have this fine tuned models. These are models that you can use to generate your own images. As I was recording the course, the DXL 0.9 model by Stability AI, it became public, now you can use that one as well. There's a bunch of different models here, for example, magic potion, spirit creatures, Christmas stickers, and so on. These are platform models, but they are also community models. And the community models were trained by Leonardo users for specific use. They made it public, so you can also try them as well. If you like any model, you can actually bookmark that model. Just click this. It's going to be in your favorite models. These are the models that I used and liked. Then there is your models here, you'll see all the models that you've trained in Nado. Okay, then we get to user tools and the first one is the AA image generation. This is where you will generate your images or prompt. Let's go back here. Then we have this AA canvas. This is where you can do in painting out painting, but you can also generate images. There is a canvas mode, you can choose the text to image to generate an image. Then we have this texture generation where you can upload a three D model and that will generate a texture from your text prompt and turn it into a three D mesh. Okay, and then for settings here, you can specify your interests, for example, art, architecture, advertising, whatever you want to see in the news feed. And then we have questions and answers, and also some guides to help you with Leonardo tools. Okay, so if we go to questions and answers. So here I want to point out one thing here. It says, can I use the images generated by the platform for commercial purposes? And it says, yes, you can use the images generated by the platform for commercial purposes. This applies to images created by free users two, which is great to know. Okay, now let's check out some pricing. Here I've got 9,500 credits. Here are all the plans offered by Leonardo. Right now, I have this apprentice plan and it gives me a lot of credit. It gives the 8,500 tokens per month. Then there is Artisan Plan which gives you 25,000 tokens per month and then Yest with 60,000 tokens per month. With free generation, currently you get 150 fast generations per day, which is quite a lot If you want just to try out and generate a few images, then with a free plan, you get 30 up scales. And with a pay plan you get more 1,000,705,000.12 thousand. Then background removal, you also get a bunch of that here. The number of jobs you can do in parallel here is only one, with apprentice 5,010 and Maestra 20. Then we have private generations. With a free plan, you cannot make any private generations, but with any other paid plan you then there is priority infrastructure. And I think that refers to the new features that Leonardo offers. Some new features would be only available to paid plan users. Then we have the relaxed generation que, which is not available in a free or apprentice plans, but it's available in artisan or master plans. Here you can check out the plan details and decide for yourself which plan you want to go with. Okay, now let's go back to the platform and let's create some art. 80. Image Generation - Text to Image: Let's generate some images. You can go directly to a image generation or you can first choose a model with which you want to generate images. And there is a bunch of featured models here. By default, when you generate images, you will be generating with this Leonardo Diffusion model, which is a proprietary model developed by Leonardo. But let's say you want to create a cartoon character, then you can use this three D animation style. Or if you're creating a game asset, then you can use that specific model. Let's try it out. Here are the featured models. If you want to see the full list, then go to the fine tuned models in this platform models. Here are all the good models here. Here, for example, we have magic potions. Now in order to create with this model, you basically just click Generate. With this model, it's also good to take note of the resolution because this was, the trading resolution was 512 by 512, it's best to use the same dimensions. Let's try it out here in image dimensions, it's already preset to 512 to 512. I don't need to adjust anything. Okay, let's maybe put something magical. I put a beautiful magic potion containing a galaxy, intricately detailed game acid illustrated. Okay, here we have the magic potion, fine tuned model. Let's out now we've got these magic images. Okay, so on the left hand side, you'll find the settings here. The first one is the number of credits that you have. Right now. I have 9,533 tokens. Then we have the number of images that are being generated every time you have your prompt. Right now I have two. Let's just increase it up to four. Then we have custom Leonardo features which are prompt magic and alchemy, and we'll talk about that a bit later. Then there is public images. If you are on the paid plan, you can generate images privately. Right now, these images are not public, it's turned off. But if you're on a free plan that it would be turned on, all your images will be public and potentially could be seen in a community feed. Okay, we can make it public. Let's say that we have the image dimensions here. You can choose the fixed or you can use the slider to change the dimension. Then we have guidance scale and step count, which are basic parameters in staple diffusion. We step counts, you may not see that as a setting depending on what schedule you choose. By default the schedule is Nado. There'll be no steps for you to manage or change. Of course, you can include a seed number. Again, basic stable diffusion parameter. We will talk about all these settings a little bit later, but right now what I want to do is to just try it out and also tell you a little bit more about prompt generation. Okay, first again, let's choose our model with which we want to generate our images. Select custom model here you can choose between your favorite models, platform models and community models. If you don't have any favorite models yet, you can go to the platform models and choose the one that you like. For example here, here we have the Dream shaper. This model is fine tuned for a portrait illustration style. That's somewhere between fort realistic and computer graphic. Let's, I actually use that one again. Let's generate with this model. Okay, for my prompt, I want something. I want a close up of a cowboy. Because this is a stable diffusion, we need a much longer prompt to generate good images For that, there is actually prompt generation tool that helps you out with your ideas for prompt generation instead of writing in this field here. You need to write in this field. Okay, so let's just copy that up short of cowboy and let's generate some ideas here. We've got some ideas. So the first one is a weathered cowboy. His face illuminated by the setting sun. His hat casting a long shadow across his rugged features. I like that one. Okay, then we have close up hands. No close up boots. His eyes squinting against the bright sun. A hint of smile playing on his lips, very poetic, maybe. Let's try the first one. You can copy that. Just copy or you just click Generate. Let's say I want to add more details here. Let's copy pasted to our held here. Now I will add some Stylize, 64 K and real engine. Okay, we already have this model chosen here. We can use the Leonato style aesthetic or no. If we want, we can also add negative prompt right now. Let's just use our prompt here. Let's click Generate. One more thing, when you use prom generation, make sure when you actually do image generation, you come back to image generation. Because when you click generate here, you would see that nothing happens. But in fact, if we go back to image generation, it already generated images of prompt. Here we've got our cowboy. Let's check it out. He looks very evil. Here I will actually regenerate and see if I like the new batch of images better. Maybe I will put the Cowboys full face eliminated by the setting sun. Okay, let's try that and maybe let's put photo realistic. Okay, let's try that. Okay, this is a bit better. I like the first one and maybe the fourth one. Okay, let's actually work with the fourth one. In the bottom here, you have different features. Here you have different up scalers. Let's try different up scalars and see how they compare. This is, the first one is Creative Upscale, and it says that it can improve images during the upscale process. It will cost us five tokens. Let's do this one simultaneously. I'll upscale with different up scalars. The next up scalar is the alternative. It says use this. If you find the creative upscale is resulting in loss detail, I guess it has more details. Let's use that one then. We have this up scalar. It works well with a focus subject but can end up smoothing out fine details. Maybe if it smooths out the background, it'd be great here, but we'll see. Then the last one is the HD up scalar. And it's a great balance up scalar which retains a good amount of detail and crispness to the image. Let's try that. Okay? Now I believe we have most of the up scalars. The first, let's go back to original image. This is the original, this is the creative upscaled image. You see how it changed certain details. Then we have the alternative upscaled image that kept most of details. And I think that's right now that's more similar to the original image. That's the alternative up scalar. Then we have the HD smooth upscaled image. Supposedly it smoothened out a few things there, and here the last one is the crisp upscaled image. I think that didn't change it, or if we zoom in on the face, maybe that has more of the skin texture. But overall, I think all the scalars are quite similar. Some added more details, maybe some smoothen out, but overall very small changes to the original image. Let's go back to the original image. Now here you have also an option to remove background, Let's do that. Here we can find the snow background image, and here we have it. The only problem, it didn't do a good job with his other shoulder, and it cleaned it up as well. That's not a desired outcome for us, but let's go back to original image then. The next feature is the zoom. That's basically out paint extends this image. Okay, let's try that. Let's see, zoomed image here, it extended the image. I think it's pretty good. The tools that we have for this zoomed image, you can copy to clipboard. You can download it, or you can delete it. Let's close it. These are basically the features that you can use. These are the features that you can use with the generated image. One thing I want to say is that with original image, if we go back, for example, to this one, let's say you have all these options here. But let's say if you upscale it and then upscale version will have only the limited settings. It will have the delete download, copy and also remove a background on the original image, will have all the features. Let's say if you don't like this upscaled image, then you need to go back to original image and then use a different upscale. That's how you do it. 81. Image Generation - Leonardo Parameters: Now let's actually experiment with the Leonardo settings here. So we have this fine tuned model dream shaper. Let's now try using the Leonardo style and see how that compares here. Okay, now for some images, we got something wrong with the hat here again, but these two are pretty similar to the default no style mode. Now let's try to generate the same prompt with Leonardo's alchemy feature. Let's turn it on. Alchemy is Leonardo's custom feature. It's designed to generate high quality two D images. It's available for paid users and you can select from a range of alchemy specific precepts. Let me show you where this is. Once you turn this on here you will see lots of different presets like anime creative dynamic photography. You can choose none. The default one will be dynamic. Also, when you turn alchemy on off On, there is a bunch of other settings that will show up such as high resolution, this boost the output resolution of lead alchemy. High resolution outputs will be somewhat different to the normal resolution outputs due to the diffusion process. It's not going to be the same as up scalers. Okay, you can turn it on or off, Let's keep it off for now. Then we have expanded domain. It increases the creativity range of generated images. When it's off, images are more likely to be aesthetically pleasing but may not be as prompt adherent. I would say that expanded domain is somewhat similar to guidance scale for alchemy feature. Let's keep it on. When we keep it on, then there'll be a greater prompt adherence, but there is the risk of visual artifacts and anomalies. Okay, we'll see if there are more artifacts, then we should turn this off. Then we have contrast boost. This will adjust the dynamic range of your image. The default is one that you may find, reducing, it is helpful depending on your subject matter. Then we have resonance that dictates how much detail is in the image and how prompt adherent it is. Around 13 to 15 is a good balance. Higher numbers will create images that are extremely busy. Right now we're on 15. Let's try out with the default ones and just see how it goes and maybe change it later. These images look way more detailed than the previous batch. Look at all the setting here and the clothing, the belts, jeans and so on. However, I think for the face, let's, I think there is something wrong with the face here. I'm not sure where are those black lines come from? For some reason they are on the other images here, it like maybe shades from the hair, but it doesn't look natural. Now, I would like to change a few settings, but before I do that, I think I'll take the seed. This generation, we have consistent composition and see what would be the differences. Okay, here to take the seed, you click on this three dots here and you can copy the seed. And then if you scroll down on the left panel and click Show Advanced Settings, You can paste the seed here and then use Fixed Seed. Turn on, Okay, now we should the same composition. Now I will change a few things. Contrast bust, it says that the default one is one. Let's try with one. Maybe it's going to give a more dynamic image. Maybe I don't want to edge detail here, I'll turn that off. Let's try again. Maybe we can also change the preset. Right now it's dynamic. Maybe we can choose a three D Render, or Illustration or Creative. Let's use dynamic since we turned on the dynamic one. We got our images and again, we have these lines on the face. I like the second one, but the previous batch was better. I really like this image here. This is not as good, especially here. We have some artifacts and again, the lines on his face that was alchemy feature, Sometimes it works really good by adding lots of details in this image, but from my experience it just adds more artifacts. It really depends on what kind of image you're trying to generate. And of course, you need to play around with all these pre settings here because for some images maybe contrast boost should be like 0.5 or zero different resonance. Or you choose to have expanded domain. That will very much depend on your image. Okay, now let's move to prompt magic. I'll turn alchemy off. You can actually keep them both on, but for simplicity, let's actually turn the alchemy off and then just keep the prompt magic on. What is prompt magic? It's Leonardo's custom under pipeline that has far greater prompt adherence, higher image fidelity, and can improve the output with any chosen model. It increases token cost due to higher GPU overheads. Once it's on, then you'll also have more settings here to change the prompt magic strength. That's how strongly prompt magic influences the output. A higher number means greater influence. Right now it's at 0.4 Let's maybe choose like the highest one here, which is 0.8 Then we also have this high contrast. And high contrast mode will give moody images with more shadows. Turn this total off if you find that outputs with it are too dark for your chosen prop, let's keep that on and just see what's going to happen. Again, I've turned the highest prompt magic strength. We see the influence of prompt magic end. Actually understand what it does here again, I'm using the same scene. Let's generate. Here we got the images in a completely different style. Here again we have cowboy, but now we have a side view. It's way more dark. His face definitely is illuminated only by the sunset. Okay, that's what we wrote in our prompt, we wrote the cowboy, where his face is illuminated by the setting sun. And I think that here and here is great compared to other ones where, for example, with alchemy doesn't seem to be like a sunset. Or maybe this one here we got the sunset. But I think these images have a more dramatic lock to it. Let's now maybe change the settings here instead of high contrast, let's turn it off and see how this will affect the image here. There's definitely less contrast in the image. Again, we have the Cowboys side view and sunset, but now the image is way more light. So it really depends on what you try to achieve. I really like these images with high contrast here. I think it really sets this Wild West atmosphere here. 82. Image Generation - SD Parameters, Schedule & Sampler: Now let's try some different prompt for this. I want to change the model. I'm going to go back and I'm going to go to fine tuned models. Right now I want to use the three D animation style model. When you click on the model, you can see the images that were generated with this model. For example, I can click more. I'll have all the images, for example I like the first one here. I can check out what prompt was used to generate this image, as well as the settings that were used. For example of the resolution guidance scale, sampler presets, prompt magic strength. We can click Remix. It will automatically use some of the settings. It will use the prompt as well as the model and maybe some of the presets. But what I found is that it may not use all of these settings, especially with prompt magic. Let's try it out. Let's use this remix button here. It changed our prompt. It added the fine tuned model three D animation style, leonato style. Actually, let's check out that image settings. Here we have resolution 640 by 832. Here we have 640 by 832. That's correct. Then we have the leonato style. Leonato style. Here they've used the pro version two, they've used high contrast, and the pro magic strength was 0.4 Let's check this out for prompt magic. It's on, but as you can see, it wasn't adjusted to that image. It was left from my cowboy. Prompt Here, I will manually change it, so I'll make it 0.4 I will turn on the high contrast. Okay, I think we've covered all the settings here. The guidance scale is 77. That's correct. Okay, let's try it out. Beautiful. We've got very similar images to the image that I found in the community feed. Okay, now I want to adjust my prompt a little bit. I want to change the setting instead of Bohemian city. I want to make it a Greek city. Here instead of Bohemian. I will change that to Tunic. In tunic and accessories. Let's try this out. The images look very nice. Okay, now I want to explain some more settings for that. I'm going to turn off the prompt magic and let's also make it private generation. Why not? Then let's repeat. What is guidance scale? Well, guidance scale is a parameter in stable diffusion. And basically it determines how much image generation process follows the text prompt. Lower values will give you more creative results, but some things from your prompt may be missing in the image. With higher values, you will have images that are better aligned with your prompt, but it may increase the risk of having artifacts. The default value is around seven. If you use somewhere 6-10 that will also give you good results depending on your image. Then we have the set number you'll find in advanced settings, and then you can use a specific seeded is just an initial input that guides the creation of the image. The same set prompt in parameters will produce the same image with minor variations. Then we have tiling. Tiling creates similar patterns. If you want to create a similar pattern, then you can turn this styling on. For example, here in the prompt, maybe let's put water color lemons here. I will disable this fixed seed. I want a random one. Let's generate, maybe not use this three D animation style. Let's just use the Leonardo Diffusion. Here we have some fun patterns that you can check in the similar pattern checker. That's the tiling parameter. Now let's turn it off, and let's move on to more advanced parameters. Another parameter in stable diffusion is called sampling method in Leonardo, it's called scheduler. Basically, it's an algorithm that guides image generation. But to be more specific, it's actually responsible for carrying out the denoising steps. If you remember how stable diffusion works, it starts with a random noise through a number of denoising steps. It creates this clean, clear image. Well, sampling method is responsible for that denoising step. There are a bunch of different sampling methods and they will produce different image outcomes. Let's look at some of them. Let's take a look. Someone has compiled this chart and shared it. On read it, it gives a really good picture of different samples. Here on the left, we have different samples. On the top here, we have a number of steps. The fewer the steps, then the noisier is the image. The more steps, the more clear and detailed does the image get. The first one is oil. I would say around step 16 we get a nice image here that we have. We get a good image at step eight, then we have LMS. As you can see, it needs more steps to get to our desired outcome that we have LMS D. Then we have Hums Sampler 02:00 P.M. two. Of course there are more different sampling methods, but these ones are the more common ones. As you can see, most of the samplers converge to the same image, except the PM two and the oiler A, which gives a completely different image. Also, as you can see, different number of steps is needed for different samplers to get to the desired image. It can be as low as eight steps. For some it needs to be 32 steps. For some it has to be more than 32 steps. It really depends on the sampling method. The fastest one is oiler. It's usually used as a default sampler for stable diffusion, but the oiler is pretty good and the DDI is also pretty fast. Now let's get back to our Leonardo AI here in the advanced settings. If we open that, we can change the scheduler or sampling method, or sampler. Here at the default on E is Leyendo. Then we have oil, or the full name is Oiler Ancestral. Oiler Discrete DIM, DPM solver, and so on. Let's try using different scheduler. For example, let's use Oiler Ancestral and keep other settings the same. I'm going to use this prompt here by clicking this button, it's going to reload you're prompt in this field, which is quite handy. Also, I want to use the three D animation model here, this one, in order to see the differences between the schedulers. I will keep the fixed seed. Let's try with oiler A. Let's also try with Leonardo. Let's with DDM. Now we've got all our images here. To see what we've generated, you can click on these three dots and view generation info. Here we have the DDM. This one is the D DIM. This one is Leonardo and this one is Oiler. You can see that there are small changes, some images here. We can see that the arm is a bit different here, but everything else very similar. The color of the dress may vary. Using different samplers will result in subtle changes in the images. 83. Image to Image Generation & ControlNet: Now let's move on to explore more settings here we can actually use image to image generation and image propped. We'll try both and we'll see the differences for image to image generation. Let's upload our image here. So let's say this one down here. I can change what's the influence of my image on the generated image. Initial strength and high initial strength will preserve the original image more. Okay, here I have the three D animation style. Let's change it to maybe Dream Shaper, because my input image is more photo realistic. I think something more realistic would work better here. Okay, Dream Shaper version seven. Let's try with initial strength of zero point. Let's use the maximum. Maybe here 0.9 And let's keep the same prompt. Okay, now let's generate here. You can see that it's almost the same as our input image here. But maybe some things were a little bit different, but not much. Maybe the highlight on her cheeks, and that's it. Let's make the initial strength smaller. Let's put 0.5 Let's generate again. Now here we have more differences. The clothing and hair accessories. Also the face looks more as the dream shape or train character. Now let's move it to maybe 0.20 0.24 at 0.24 Now the background is also different. See how the strength of 0.5 the background was kept as on the input image, but with the 0.24 Now we use more of the prompt here. It says Greek City Landmark and we get those columns. Okay, that was the image to image. Let's see how image prompt is different. First of all, image prompt here, you can upload more than one image. For example, here we have this lady here, but I also want to give a background here. I have a Greek town, Santorini. Here let me show you, this is the input image of the lady and this is a picture somewhere in Santorini. I'm putting all those two images here. I have different settings. Here I have image weight and higher values will make the output look more like the reference images. Low values adhere more to the text. I think the 0.7 the default one is pretty good. Then we have this magic string which is bad custom function. Here it says how strongly pro magic influences the output. A higher number means greater influence. Let's keep it in default 0.4 we can make it the minimum is 0.1 and the maximum is 0.8 Yeah, let's keep the default 10.4 Let's see here, we definitely see the influence of the input images On the background, we have the Santorini City, this blue roof in a dome shape. Here in the foreground, we have this lady and her posture is quite similar to the input image here. The difference again, between the image prompt and the image to image is that with image prompt, you can upload many images as well. Use the prompted magic with image to image, you cannot use prompt magic. Now let's move on to our next setting parameter, which is control net. I think there is. Good transition between using the image to image and image prompt. Let me explain. When you upload image here, there is no way you can tell AI. What do you want from this image? Let's say I only want to use the posture of this lady here. I don't care about the background or anything, just the posture or maybe a facial expression. But with image to image, this is very difficult to do because all you can do is affect the image strength, which is the image weight. And you don't know what part of the image will be used in the image generation. For that reason, there is a very useful tool called control net. Control net is extension or add on to stable diffusion that allows to input a reference image to influence a specific attribute of the generated image. It can include a pause, composition, edges, depth, or facial expression. Let me show you, these are different control net models and this is the image input. Different models extract different things from our image input. The first model is called in Leonardo, is called edge to image. What can does, let me show you the candy method extracts the hard edges of the sample image. It's useful for many different types of images, specifically where you want to preserve small details and the general look of an image. The image input was this meme here. Here, what it extracted here is the image that was generated with. Then we have our depth model. The depth method extracts the three D elements of the sample image. It is best suited for complex environments and general composition. Let's check out, see how we have this white in the foreground and a little bit more gray in middle ground and black in the background. Here is the image that was with the depth method. Then we have open pose, and the open post method extracts the human poses of the sample image. It helps tremendously to get the desired shot and composition of your generated characters. Let's check out here, this is what it extracted from the input image. And see how here and now we have totally different background, totally different clothing, but the poses are the same. There are more models for control net, such as Scribble and so on. But for Leonardo, right now we have those three models. If we go to Leonardo, let's switch on control net. Here we can choose post to image, edge to image, and depth to image. Let's try all of them. Here I have the post to image. Make sure you have the image here. As you can see, now there is no initial strength here. We can only change the control net weight. The higher the weight, the more control net will influence the generation output. I think one would work best here because I want to have this exact pose. Let's try it out. Again, I'm using the same prompt. Let's generate, see how we got exactly the same pose, but now the background is different. The woman here is very different. Clothing items are diverse compare to image generation, where we have almost the same lighting to the original image. If the background is different, then not too much from the original image, but here completely different lighting. Only the pose is what has been taken from the original image. The control net gives us so much more control over image to image generation. Now here, let's check out those images first to make sure the proportions are wrong. This one is way better here. This one is good too. Here again, proportions are a bit screwed up. Let's try, instead of post to image, let's use edge to image. It's going to use all the edges from the image here. See how the face is more similar to the original image then compared to the post to image here face is completely different. Then we still have the same hose, but the background is different. And that's because here in the edge to image, we have very detailed information about the girl here, about her facial features. Including her facial features, but for the background, because this image has blurry background. Maybe some small things here with edge to image. Let's say if this background wasn't blurry and we had some nice and sharp buildings in the background, then it would also catch that and will generate something that would have the same shape. Now let's drive our last model, depth to image. Let's try it out to remind you the depth to image model extracts the three D elements from the input image. Now here we got this girl and she is in the same pose as our input image. The face is quite different as well as the background is also completely. For the depth to image model, I think it'll be more fun to have more characters on the image. Let's maybe change the image here. I will app different image, let me show you here is the image that I've applauded. And here we have a girl in a foreground reading a book. On the middle ground we have a different girl. Let's try that one. I will here change the prompt a little bit as well. Beautiful woman in tunic. Let's put sitting on a bench. Let's keep everything else the same here. I didn't pay attention. When I re appload a different image, the control net automatically turns off. This is just the simple image to image generation, let's just check that out. Here are the images. Now let's turn on the control net and use the depth to image. And we'll see how that compares. See how in these images, the position and the relative position between these two characters is the same as the input image compared to the image and image generation where the correct position wasn't captured. That was the depth to image. You can see that there is a huge difference between using control net and just the regular image to image generation. As you can see, Leonato has so many settings that you can adjust to get a desired outcome image. Here we've talked about leads, custom functions, prompt magic and alchemy. We've talked about switching between public and private generation, where you can change image dimensions, you can change the guidance scale. Then if you have a reference image, then you can use the control net with three different models, Post to image, edge to image, and depth to image. Then we have tiling, Then you can use the regular image to image generation. You can use Image Prompt with many images. Then you can use a fixed seed. Then you can change the scheduler, which is the same name as sampling method or sampler. 84. AI Canvas - Outpainting: Now I would like to move on to A, A canvas. This is where you can do in painting and out painting. First of all, here in the canvas mode, you can choose the mode. The first one is text to image. Let's say you want to generate image first, then you can use this one here. Here, you'll have most of the settings that we've already seen in the image generator settings. However, let's say you have generated an image. Let's actually go back here and let's go to image generation Here. I want to find a nice image. Maybe this one can actually send it to the canvas by clicking the square icon. Now it's in our canvas and we can work with it here. Okay, for now, let's change our canvas mode. Since we're not planning to generate any new images, let's use in paint and out paint. On the left hand panel here we have a pan tool that basically allows us to move the canvas around. Then we have this Select tool. This allows us to select this generation frame and move that around. Then we have Draw Mask and Erase Tools as tool. Let's say you want to change something in this image, you can go ahead and erase that. For draw mask, it's very similar. If you want to change something about this image, you just paint over that specific element. What's the difference between rays and draw mask tools? Is that the mask retains some information of the image under the mask. For example, if you want to draw sunglasses and you want to retain maybe the eyes here, then you can use the mask instead of erase tool. Then we have the sketch tool that allows you to draw something. You can draw it on the canvas here or you can draw it on the image on the top here. You can change the colors. You can change the brush size as well. Lastly, we have this upload image. You can upload images from your computer. You can download the artwork. Okay, now let's say I want to make out painting. I want to extend the image. Let's extend it maybe to the bottom here. I will use the Pan tool to move the image bit up. Then I will use the Select tool to move my generation frame. Down here on the right. I have a bunch of other settings here if you want to out paint, make sure this out paint is turned on. If you want to do in painting, then you can turn that off and then you can choose the paint string. But we'll get to that too for now. Let's use the out paint then How many images that you want to generate for the out painting? Let's use four. Then we can set image dimension it's nice to generate with highest size. So I'll change that and see how our generation frame immediately became large. Then we have this render density. Render density decreases the size and increases the pixel density of the generation context. It will decrease the size of this generation frame, but the quality of the image that it will produce will be better. Actually, right now it's one X, maybe. Let's make it 1.5 or two X. See now the size is smaller but the image generate will be higher quality. Then again, we can change the guidance scale. And let's see some more advanced settings here. We can use the fixed seed and we can choose the scheduler for now. Let's keep that as default here, I will use the same prompt as I used to generate this image. Also make sure that the generation frame overlaps with your image here. May move bit up here. For negative prompt, you can also include that by clicking this button here. Let's say forward. And nude. Let's click Generate. Let's check out the images. This one is missing a hand here. We got completely wrong extensions for that reason. Let's actually cancel it to improve the out painting results. There are a couple of things that we can change here. If you've noticed to do the first out painting, we've used stable diffusion 1.5 model. But when I was generating this image, I actually used a fine tuned model called Dream Shap. Let's switch and see if that will improve the out painting results. Dream Shaper seven, that's the one I've used here. I will keep the same prompt. Let's generate, let's check out the results here. Here, again, we're missing the hand here, but I think it's a little bit better matching. Let's see the other ones. Okay, still there is a problem with the arm, but I think the details match way better here As we've changed the model, let's cancel it and help AI a little bit more. I will erase the sharp edge here. I will remove some of the back here and the arm, hopefully that will give it more room to improve. Let's generated again here. Now we have, I think, a better extension of this image. The arm is covered with this fabric, which is not too bad. Let's see the other ones. Yes, same here. This has a bit more problem. Okay, I think the first one was the best. But we can spend a little bit more time here and just try changing a few things in the prompt and in the settings to make it look even better. Let's do that after playing around with the settings. Here is the result that I've got pretty like this image here. This is where I stopped. I've changed the size of the generation frame. I reduced the render density to make this generation frame larger so it covers more of the area. Also, I changed the prompt here a little bit. I added with bracelet on her hand, This is not really reflected here. But then I also increase the guidance scale to eight, which is also a minor change. Then I just try to generate a number of times to get the result that I want. Here are some other images that I've got, but they weren't as good as this one. As you can see, there is still problem with the hand here. But I like how it's covered here. Right now here, it just looks natural like she is covered by this fabric. I will click except here. Now we can also add maybe some out painting to the sides. I will keep this a Greek city landmark background and generate a couple of frames here. Again, I will move my generation frame to the right here. I will use the array tool to remove these edges. Let's this looks good, so I'll just accept it here. So this is a final result after all, the out painting. 85. AI Canvas - Inpainting: Now let's try in painting to switch on to paint. Just turn off this out pin tool here. Now we can change the pain strength. The painting algorithm looks at what already exists in your image to better edit and place new quantum. The higher the strength, the more it will diverge from the original image. The default one is one that's the highest. Let's say if you want to keep some of the content, then you can reduce the in pain strap. But let's strive with the default one here. Now I'm going to zoom in. I want to add a tattoo for that, I'm going to use all the tools here. I'll use the mask, I'll use a race, and I'll use the sketch. So you can see the differences here. I will start with, let's use the rays one first. On the top, I can change the size of my brush, make it smaller. Let's put 20. Okay, I want something here and here in the prompt, I will pour a butterfly tattoo. Okay. So now my pain strength is one. Image dimensions is 1020, 4,024 Rented density is 1.5 x. We can actually make it higher because the area is very small. Let's make it maybe 2.5 x or even higher, 33.5 I think that's good. Okay, again, let's zoom in and use the pen tool to move the canvas in. Let's generate. We have this butterfly tattoo. I'll lick the blue color here. I think this one is the best, because it matches better with this whole image. Let's accept it. Now, let me show you how you can use the sketch tool to generate something that you paint. For example, here I will draw cherries. Here, I will use the red color. By the way, this is a color picker. For example, you want to use this color here. And you just can click on this button and click on the color you want. Now see a match to the color of this place here. Let's change it to red. Think this is a good size. Now I will draw my cherry. I have one, maybe a second one, and I want a stamp. Let's make it green. And let's decrease the brush size. 14, I think that's good. Or even smaller. Put eight, that's good. Here I'll put and a leaf something like that at the bottom. I'll put a cherry tattoo. Now instead of using this in Pain out Pain tool, I will go to the Canvas mode. I will go to Sketch. To Image, I will choose the mode. Once the mode is selected, I will click January. Here we got the images that look like my sketch. Let's see other variations. Okay, again we have those two cherries and leaves. I like the first one here. Let's accept it. Okay, now let's use this mask tool here. I will actually cover this butterfly here. Just to show you what the mask does. Maybe we will cover a little bit more area here. I'll put butterflies tattoo again. I'll switch the mode from sketch to image to paint and out paint here. Let's say we want to preserve some of this butterfly or maybe the shape, then we can make this paint strength smaller, make it maybe 0.8 Let's use that. Let's generate here in place of that old butterfly tattoo. We got these new butterflies. As you can see, the shape of the butterfly and the colors are quite similar to the previous tattoo. Let's see the other ones, again, very similar shape like this one the most. This is how you can use the arrays tool and then the mask tool to change things around a bit. I'll also like how the inside of this butterfly match with her skin tone. It really looks pretty. Let's accept it. Okay, I think this is way better here. Now let's continue and do some more in painting. I will zoom out a little bit to show you the final result. I think the butterfly looks really good here. But for the char is, maybe we should change. But that's for later. Okay, now I'll use the Pan tool to move the canvas. And I'll use the Select tool to move my generation frame. Again, I'll zoom in here. I want to add sunglasses. I'm focusing my frame around her eyes. I can make it even smaller, so maybe four X. I think that's good. Here and here I can choose either to draw a mask or erase, or even sketch. Let's use the mask and I will change the brush here in the prompt output. Fashionable sunglasses. Let's generate the glasses are poorly integrated. Here we have big lines. Let's see the others. This is a bit better. I like how we try to preserve the eyes. Let's generate one more time and see. I think these glasses turned out pretty good, but again, we have this problem here. But we can probably fix it with image to image generation. And let me show you how to do that. Here I will click, except now I will change the canvas mode to image to image. Here, prompt, I will put my old prompt, Master Professional Photography and then beautiful woman in tunic. Here I will delete the tunic and put it in fashionable sunglasses. Let's try it out, by the way, for input strength here, we can also modify this. That's how closely the generation will follow the underlying image. Let's say if you want just small adjustments to the image, then you should make it higher. In our case, it's 0.3 Let's move it higher because I don't want to see a big difference. Maybe let's move it to 0.5 If we have strange results, we'll move it even higher here. As you can see, image to image changed the whole area in this generation frame. Now we see a completely different phase. Let's see the other ones. This one is pretty good. I like the earrings here. The sunglasses now look way better. Maybe the second one was pretty good. Now let's actually cancel that. Let's make the input strength higher and see what will happen. Let's cancel this. Let's make input strength to the highest is 0.8 Let's make it 0.65 or 0.7 let's make it 0.7 Here we see just slight modifications but, but it makes the whole image way better, especially the sun glasses here. Let's see other ones. I think the first one was pretty good, but the only problem is that we don't have this frame here. Actually, let's cancel it, and I'll make the input string for a little bit less 0.6 Let's generate again, let's check the other ones. This one tally has artifacts. Well, I like this one. I'll keep this one with image to image. We turned this to this, which I think is a pretty good result. Okay, now the last step here is to download our artwork. Let's check it out. Okay, so this is our final art work. Okay, with image to image, I can definitely see that we've got this square a little bit out of place. You definitely can see those different pixels here on the canvas. We can actually blend that together using the array tool. I will the edges here. I will make my generation frame higher. I'll reduce the ranger density. That should work. Before I click Generate, make sure to change the mode to paint out paint. Otherwise we'll just get the image to image generation, which I don't want here. Let's use the paint out paint now. As I'm looking at it here, it nicely blended it together. Let's see the other ones, there's like slight changes in the shoulder. I like the last one here I'll accept. And again, see the difference here. Now it's way better. Let's download it again. This is our file Result with Out painting and in painting using Leonato's AI canvas. 86. Leonardo - Training a Model: In the introduction to Leonardo, I've told you that you can actually trade your own model and you can do it here in the training and dataset if you have a free plan. I believe you can train up to one model. But if you are on a paid plan, there are more models that you can train here. You can train your character here. I trained with myself. I trained with a character, I also trained with a style sticker style. Okay, how can you train the dataset? Basically, just click new dataset here. Give a name to your dataset, for example, fun stickers. And then you can give a description for your collection. For example, animal stickers. Then let's create a dataset here. All you need is to drag and drop the images. For example, here I have a few stickers. Let's put them all here. Once you upload all the images that you want, you can click this train model. But before we click this train model button, let's go and check out the requirements and tips for the best model training. This you can find in if you go to the questions and answers help and AI Model Fine tuning. Introduction to model fine tuning. Here is some steps. Here we have what makes a good training number of images. So suggested ideal eight to 15 images, minimum five images, maximum 30 images. Here is more suggestions on variation and consistency. Consistency, It's important that there is a common theme or pattern between your images for the model to learn from. For variation, things that vary across your images will be more loosely learned. That is what allows your model to put your trained object in new kinds of style and context. Let's check some examples. These are bad datasets. As you can see, these are very different styles here. And then we have also repeated images. This is a good data set. We have animals that are quite similar style, different backgrounds, different animals. But that's a healthy variation that helps the model to learn. Let's go back to our training and datasets here again, let's applaud more images. I will applaud all of them. These images I generated with Leonato beforehand. Here we have 13 images. I think they're quite diverse and yet they are in a similar sticker style. Now all we need is to click this train model here. Model name. Let's put stickers then, training resolution. One more thing, when you upload images, make sure that they are squares, otherwise you may get very strange results. Let's go back to our train model, a fun stickers for training resolution. You can choose 512 or 768 by 768 per category. Make sure that it aligns with the characters so it can be a character environment building, fashion illustration photography, if you're doing maybe photos of yourself, for example, that we have product design textures, elements in this one is set somewhere between illustration and graphical elements. Let's use illustration. Then we choose the model stable diffusion, 1.5 or 2.1 I would make sure that the training resolution is consistent between the model. Here you can see 512, 512 here. If you choose 768, then you should probably use the version 2.1 However, it also says here that stable diffusion 1.5 is recommended. This model performs well in general, better with realistic images. Okay, model description animals. I guess one of the most important things here is the instance prompt. This is what you will need to put in your prompt to refer to your model or your character. For example, if you applaud selfies of yourself, you can perhaps put, I don't know your name. Then in your prompt you'll put and then a description such as a girl with curly hair in walking in a park, something like that. Here for example, you can use three letter combinations such as SKS. It doesn't quite matter. Just try not to put something vague here. I'll just put SKS. I'll just start training right now. Training is in progress and it says that they will e mail me when it's complete. Depending on the size of your training model, it may take anywhere between 30 minutes to a few hours. Then you can check the status. But I've trained my stickers. Let's just try it out and see how it works. Again, I will go to AI Image generation here for example, I can put sticker, cartoon cute baby husky, white background, 12 K, high quality HD octane render. Then it's important to choose your model here. You can select custom model in your models. After training, you'll see all the models here. For example, let's choose stickers to just click View and click Generate with this model. Okay, now we have our model selected. And very important, make sure that in your prompt, it doesn't matter where, but you put the instance prompt that you've created. In my case it's SS sticker. And I have to add that so that the model will be generating based on my training images. S sticker, cartoon cute baby sky white background 12, high quality H C octane render. Now here I don't want to see Lana style and I will disable the magic. Let's also make it private because it's based on my model. Then for image dimensions, I've trained it on 768 by 768, this is good. Then the guidance scale, then let's check out the scheduler. Here we have Leonardo. I will actually change it to Oiler Ancestral Steps. Let's use the 30 steps. Okay, let's check this out. I think the results are successful, except for the eyes. I don't like them here, but everything else is very similar to my training data set. We have this sticker, white highlight here and grayish background. If you notice all my images here, they also have grayish background. Okay, that was for the stickers. We can try something else. Maybe not husky, but maybe bulldog, and let's try that. Okay, this is so cute. In a similar way, if you have a specific style that you like working with, you can train a model on that style. And then I use this model to something else, something new. Let's now also try with myself. Here I've created a model called T, which was trained on myself. Here I will click on, I'll click Generate with this model. For the prompt I will, my instance prompt here is a girl. I have to include that here. A girl with curly hair in a part background. And then I'll put soft lighting, high quality, and let's put 64 K for negative prompt. I will also add that. Bad framing and so on, Okay. Now. Here. It was trained with 512 by 512, so the image dimensions is correct. Okay, so let's increase the step count a little bit here. Movie 30, 40. I think that's good. Let's generate. Okay, so that me or the AI generated version of me overall, it's not bad. It captured my hair style, it captured my facial features. My eyebrows, nose, eyes, mouth. However, the quality is not great. We can try to upscale. Let's use maybe creative up scalar after upscaling with all the up scalars. Here are the results. This is the creative up scalar. This is the alternative. This one is the HD smooth. This is the H D crisp up scalar. The best result here was the original image without any up scalars. Because the upscalithermooved out too much, they changed the facial features and now it just looks ugly. The original image was the best here, but again, unfortunately the quality is the best here and up scalers do not help. Of course, we can change and play around with different prompts here and change the settings to play around with guidance scales, step counts, schedulers, and so on. Another thing I want to note is that if we go back to our training and datasets here, we have quite limited training settings. The advanced training settings are not available yet, probably that will definitely change in the future. But for now it's very simple, and for people that want more control over the training, this is very limited. Another limitation here is that when the training is finalized, you cannot download the model and use it with your own stable diffusion. But on the other hand, Leonardo allows to train one model on a free plan, which is a big advantage, and you can try it out and have fun with generating images based on your model. In the next module, we will talk more about model training. And we will actually cover another platform where you can train your model. But right now, let's quickly sum up what we've covered in Leonardo. Well, first we started with home page. Here you'll find the featured models. And we've talked about the gallery. We've also talked about the community feed and the personal feed. Then we went on exploring fine tuned models and using some of that in AA image generation. Then we went on to generate images and we generated a bunch of different characters, images, and other art. Here we've covered all the settings. On the left, we've talked about image generation and prompt generation. We've talked about prompt and negative. Prompt, fine tuned model and different styles. Then we went on exploring the AA canvas here. We've talked about all the tools here. On the left, we've changed the models as well as we looked at different canvas modes, such as in Pines, out panes, image to image, and then sketch to image. We've also tried different settings here as well. Then we went on to talk about training your own model and generating images with that model on this node. I want to finish with Granada module and let you try it yourself. 87. Astria.ai Introduction: Hello, hello. In this module, we will talk about fine tuning stable diffusion models. And we will cover a platform called Asta. Let's check it out. Asi was founded in December 2022. It's a website where you can do high quality fine tuning and image generation with stable diffusion. Here you can train custom models, download them, and generate images using your model or some other stable diffusion models as well. It has very simple interface and it's easy to use. Before we go on and check out Astra, do I, I want to talk a little bit more about fine tuning and what is it. Fine tuned model is a model that's been trained on a specific data set such as a particular art style objects, animals, people. How does it work? Well, a model that's been trained on a wide data set, such as table diffusion version 1.5 which is a general or standard model, is now further trained on a narrower dataset. Why do we need it? Why do we care about fine tuned model? Well, general models are good, but they cannot be good at everything. For example, if you want to generate images in a particular style, such as a specific anime style, it can be very challenging to do with the prompt. Fine tuning allows us to train the model in that art style. When we generate images, we will get them consistently in that style. The fine tuned model will be biased towards generating images similar to your dataset while keeping versatility of the general model. There are a number of fine tuned techniques, one of them is Dream Booth and another one is Laura Dream Boof Technique was developed by Google and Boston University. And it involves injecting a custom subject into the general model. There are many platforms that help you train models using this technique. And this includes Astra. You could also run Dream Booth yourself on a computer or using a cloud service like Google Coap. But what's the advantage of using platforms like Asta is that they already have specific presets that allow you to train your model in just a few clicks. And the results can be way higher quality than personal dream booth training, especially if you're a beginner. That's why using services such as Astra.ai, is a good starting point in fine tuning your models. Let's get started and let me show you Astrid. Ai. Here is Astra's website, Atria. Let's check out the pricing first. I'll click here. Asta doesn't have any free credits or free trial pay as you go in order to train your fine tune your model. It's going to be 1.5 dollar per model. If you want to generate images with your model, then it's $0.10 per prompt. So you'll generate eight images. Let's say if you want to make video with Asta, that's going to be $0.40 per 100 Phrase here, below are some more details. The minimum credit card processing amount is $5 so you cannot put anywhere less than five. Also, models are only saved for 30 days since the moment they become available. However, you can extend model storage and it's going to be $0.50 per model per month. Then also new is not allowed and so on. Then at the top here, we have community gallery. We have lots of community feed. Let's check out the gallery. Maybe here first. Okay. Apparently, before viewing the gallery, you have to sign up. Or again, if you don't have Atrias account, then sign up. It's very easy to do. Just provide email password here. I will again. Okay. Now I'm logged in with my e mail here and let's go to check out the gallery again. In the gallery, you'll find prompts that were made by the astray. I quite like it because as you can see, the prompts are pretty lengthy and they contain also the weights and the negative prompt, which makes the result really good. You can use these prompts when you generate images with your own model and take that as an inspiration. Then there models, these models were fine tuned and made available by the community, so you can also check them out and use them. Then there is video API and examples. Let's check examples. Here you will find what things can be trained with the model. For example, a car here is the original image it was trained and after. Here are the images that were generated pretty good, then there is a hat. Here are the generated images, concepts, different style genres here. And then we have a man also very different or techniques. Then we go to tunes. This is where you'll find all your models. This is my model that I've generated beforehand with the character. Then we'll go to, here is the AA image generator. Here where you write your prompt. Then you choose your model. You can use this basic interface and concrete image. Or you can use the advanced interface where you can add the negative prop change parameters and so on. So this is the overview of Astra's website. In the next video, we will go on fine tuning our model and also trying out to generate images here. 88. Astria - Training a Model: Now let's fine tune our model. Go to tunes. Once you find your account, you can fine tune a new model. Just click this new fine tune here. Give a name to your model such as a. This doesn't matter too much. Then very important, write a class name for your model. It has to align with the subject of your training dataset. Because I'm training this model on myself, I'm going to put a woman, but you can also put like a couple man book, dog, car hat style and so on. Let's use woman here then I'll choose my images. Here are myself here, it says upload 20 images of the subject or anywhere 4-30 images should include subject in different variations, expressions, poses, and backgrounds. Go to your photos. Up and select photos from different types. No nude here. If we scroll down here, there is also a little bit more information here. The images preferably should be cropped to one to one aspect ratio, two square. Then they recommend uploading three photos or full body or entire object. Five medium shot photos from chest up and ten close ups. Then variation is key. Change, body post every picture, use pictures from different days. Backgrounds and lighting show a variety of expressions and emotions. Make sure you capture the subject's eyes looking in different directions for different images. Take one with closed eyes. Every picture of your subject should introduce new info about your subject. Here we've got more details on the generation process here. The fine tune should take around 20 minutes. It can take longer depending on the cue. The models will be saved for 30 days. But there is an extend option. Okay, let's try it out. Here is the basic interface. Can click this Creed. If you want a little bit more control on how you train your model, you can click this advanced button. And now we have more advanced features here. For example, we can select which model we want to use here. We can choose between stable diffusion and 1.52 0.1 open journey two and so on. So here we have a general model, but you can also select a base fine tuned model. It may sound strange, but you can fine tune a fine tuned model Here for example, there is a model that's been trained on a specific style, like let's say photography style. It's still broad enough to train your own subjects with this model, but it's going to give you a specific aesthetic. For example, here we have CB I and I can't believe it's not photography here. If you just Google this model, you can click on this AI. Here is a little bit more information about the model and you can see that the images that are generated with this model are very photo realistic. Let's say you want to train your portraits with this model, you can just make sure, let's close this here. Make sure that the base model here, we have stable division 1.5 is aligned with this model here. So here we've chosen stable division 1.5 and from the information here, the base model is also stable division 1.5 It matches, otherwise you'll get strange results. Okay, now here you can write your own token. This is the same as Instance prompt. This is what you'll put in the prompt to refer to your subject. It can be SKS, it can be whatever. Here, let's use S here. You can also choose the model type. Here is the Dream Booth technique or Laura. Laura right now is in the beta mode right now. Let's keep it in Dream Booth, Fine tuning. Okay, now we have all the information here, The class name, we've uploaded our images. Let's create it. It's now going to upload all my images. The fine tuning has begun. It says here the ETA is around 30 minutes. After waiting about half an hour, our fine tuning is complete now, and this is what you will see. You'll have generated images with this new model using sample prompts. The first one is the '80s portrait of SKS, woman, blond hair, blue eyes. Then we have realistic digital painting here. I think the facial features are, don't quite match, but maybe this one is not bad. I like the number three here. We have magazine cover photo realistic glamor shot of beautiful SKS woman here. So far I like it the most in terms of a facial feature similarity. I think this is a spot on. I really like this one. Then we have number four, close up or face of SKS, woman fashion model. Portrait of young SGS woman, cinematic flower patterns and so on. And the last one is a portrait of Princess S woman. If you like any of the images, you can download all of them here. If you want to download all of the images here, you can click Download Generations. And this will download all of the images here. Let's do that. Okay, here, let's put cat generated images here. You'll have a Si folder with all the images here. Okay. You can also write your own prompt here. Just write your prompt then. Let's say you want to also add a negative prompt. You can click on this Advanced button. Not only you can add your negative prompt, but you can also configure other settings like Steps, See then we have with it. But right now, let's keep it more simple and just use the prompt and negative prompt. For inspiration, I will go to the gallery and choose some of my favorite prompts there. Here I can filter out that we just see women images woman. And now here let's choose the best prompt. This is like retro style. I think the first one is pretty fun. I'll use that. I'll copy all of it here. And also I'll copy the negative prompt. Here is my prompt. Okay, and then for negative prompt, let's also copy it. Okay, here, maybe let's reduce the number of images instead of 88 is default. Let's use four here and let's create image. One more thing that I forgot to mention is that when you copy and paste the prompt here and you generate with your own model, make sure that the prompt includes the instance prompt here we were that they've used the same instance prompt as, um, I created with my model. Here I'm using SKS woman as my instance prompt and they also used it here. One more thing, when we were training our model for the token, we just put the SKS. But here it's also important to add the class name. For the class name, we put a woman. It's best to include both of the token and the classname SKS woman. Make sure that you added in your prompt, otherwise the images that you'll generate will not look anything like the subject. Okay, so let's reload and check it out. Here are the results. I like the results way more than the leads model training. I guess that's because Astra is specialized in model training. They have better chosen settings for Dream Booth, Fine tuning. If you have more than one model with them, you can check all of your models in the tunes. Here I have my models, the new one with Cat and the older one with AI avatar. If you notice here it says deleted after 30 days the model will get deleted and you cannot use it for image generation. Actually, let me show you if I click on this model here you can see that there is no option for prompt writing because the model was deleted and I cannot do anything. Let's go back for T. If I click here, you can see that I can write my prompt here and negative prompt and so on. I can generate images with this model. Also, if I go to Gen here, I can write my prom automatically. My model will be pre selected. If I have one however, I can change it. I can choose from any of these base or fine tuned models here, but if you have more than one model, you can easily go to generate and choose between your models. But you can see that my other model, the AA avatar, is not here. And that's again because it's been deleted. What if you don't want the model to be deleted after 30 days? Well, there are two options. First you can extend the model storage. For that, go to your account here in the billing here, I can turn this on automatically extend model storage and it's going to be $0.50 per month per fine tune, and it's going to be deducted from your balance. Make sure you have a sufficient balance in your account so your model doesn't get deleted. Another way you can do, let's say you don't want to extend the model storage here. If you go to tunes, click on the model here, you can actually download the model. Just click on this KPT file and this will download the whole model. It's going to take some time because usually the models are quite heavy. This one is 2 gigabytes and it's going to take around 40 minutes to download. After downloading the model, this is the file that you'll have, it's going to be with a CPT extension. Once you have your model, you can use it with stable diffusion that you run locally on your computer or on any cloud server, such as Google Colab. And in the next model, we will precisely cover that. I will show you how to run stable diffusion yourself and how you can use this model that we've just downloaded with it. See you then. 89. AUTOMATIC1111 Introduction: Hello everyone. In this module we will cover our last A, A application which is Automatic 11 11. Or the abbreviation is 11 11. What is it? It's a popular web interface for running stable diffusion for advanced users. It was developed by a user with a nickname Automatic 11 11 with contributions from a passionate community. Here you can run, train, and deploy stable diffusion models. It constantly is being improved and updated by the community on Github. So you'll get up to date features and the newest extensions. Let's check out some pros and columns for automatic 11, 11 for the pros. Here you have high image customization and control with lots of features and settings that you can adjust. Yes, as I already said, there are many parameters and settings to achieve your desired results. The quality of images depends on your expertise. If you're good with stable diffusion, you can create really beautiful images. Automatic 11 11 has a lot of features which include text to image, image to image in painting out painting. Then upscaling base recovery and lots more. It also supports many extensions and add ons for AA art and video creation. We've already talked about control Net in Leonardo's module. Now you're familiar with that. This is one of the extensions for automatic 11, 11. If you want to create videos, then you can use to form and there's lots more extensions that you can use with it. Another big advantage of Automatic 11, 11 is that you can use your custom models or choose from a wide variety of models. So you can choose from thousands of models that were made public by community and try them out with Automatic 11, 11. Also, when you run Automatic 11 11 locally on your computer, there are no restrictions in terms of what kind of images you can generate or what kind of prompts you can use. If you remember from images they were flagging specific words. Here are no filters or censorship, so you can generate any image that you want. Now, let's talk about cons. So the first one is that Automatic 11 11 has an advanced user interface with so many features that it's going to be very overwhelming for a beginner. Second of all, if you want to install automatic 11 11 on your computer, it can be quite challenging to do so because we've already got used to the softwares where you download them and you just click Run and they automatically install. However, here you would have to use Terminal on Mac or Command Prompt on PC. If you're not familiar with that, then it might be a bit challenging for you. Also, when you install automatic 11 11 on your computer, it requires a powerful GPU. Not all the computers qualify. If you use a C, then you can use M1m2. Finally, it takes experience to generate high quality images. If you are a beginner, then maybe using more beginner friendly platforms like Lexica will be easier and you will just get better results. But right now, with all the material that we've covered in the course, you should be well equipped to generate some grade images in Automatic 11, 11. Now let's take a look on how you can set up automatic 11, 11. There are different options here. You can run automatic 11 11 locally on your computer. Here are some installation guidelines in this course, I'm not going to show you the installation process because everybody has different operating systems, so it will be quite different for everyone. Another option you have is to run automatic 11 11 on a cloud server. That's definitely helpful if your computer does not meet the requirements for installation. In the next video, I will show you how to set up Automatic 11 11 on Google Colab. Here are useful links. Also, I'll be showing you how to set it up in run diffusion. I also want to mention that when you install and run Automatic 11 11 locally, then it's going to be free. But if you use Cloud Server with Google Coll Ap, you may need to pay round Diffusion is a paid service. This is also something that you need to take into the account when choosing between the different setups. This is all for Automatic. 11 11 introduction. In the next few videos, we will set up automatic as well as Get started and I will show you all the different features there. 90. AUTOMATIC1111 Google Collaboration Setup: Let's begin the set up with Google Colab. All you need is just click this link here. That's going to take you to Google Colab in the beginning. Here we have instructions, updates. If you want more information, you can check out this guide on how to use this notebook. Just click on the link. Okay, let's get started. It's not too hard. Here we have username password you can change you want, but let's keep it as default. First, here is the option to save small model images and settings in your Google Drive. If you want to do that, you can keep that as default. If you don't want to save anything, then click nothing. If you're a more advanced user and you want to save everything, then choose this one. But it's just going to take a lot of space from your Google Drive. Let's just keep the small images and settings here, then here we can choose what models we want to use with Automatic 11 11. Here we have a version 1.5 the 1.4 F222 model. This one is nice. The Dream Shaper model is also good. There are a bunch of models that you can choose from. You can also select the version two models. Version two with 768 by 768 train dataset and so on. Okay, for now let's use the default 1.5 model. Then we have extensions. So if you want to add control net, then you just click that control net here. If you want to use different models that are not listed here, then you can provide a URL with that model. And the same applies to extensions. If let's say there is a different extension that you want to use that's not listed here, then you just use the URL and paste it here. Okay, so right now we basically didn't change anything except I just added the control net. Now let's run it. Just click this Play button here. Okay. And here we need to click continue Anyway, because we want to save images and small models on Google Drive, it asks to connect to Google Drive. Let's give the access. Now. It's going to take some time to launch automatic and run all of this code here, install everything that it needs to work. Right now, we just need to wait. After around 6 minutes, I've got this local and public URL and that's how you know that it's good and running. However, if you find that it's been quite a while and still loading, if you've clicked lots of models and extensions, try to reduce that. Also, if you want to use heavy extensions like Control Net, then you may consider upgrading your Google Collapse. You can try table diffusion with the free Google Coap first. But if you find that you need a little bit more, there are the different plans here. For example, Coa gives you 100 compute units per month, that means faster GPU and more memory. Also there is coal, which gives you 500 compute units per month, Faster GPU, more memory, background execution terminal, all the good stuff here. We can just click on this link here. And it's going to open in a new tab here for username and password. Put the same username and password that you've entered here. The default one is a, let's use that. Here is our web interface to stable diffusion. Here on the top, we can choose the model if you selected more models that you can choose from all the different models here. Here I only have the version 1.5 Here is where you write your prompt, the negative prompt, and click Generate. Okay, let's try some of our prompts here. I will use the first prompt, Professional Portrait Photograph. Then I'll put the negative prompt here. You have default settings, you don't have to change here anything. Let's generate first, here is our result. It's cropped, okay. I don't think we have that in our negative from Let's put cropped also. Let's increase the badge count is how many images you want to see being generated in parallel gills. Again, try that once you've generated these images. If in this step you didn't change anything and you kept save in Google Drive, small models, images and settings, you will find the images in your Google Drive. So if I go to my Google Drive here, I can see the output. You will see a folder called AI picks. If you click on this, here will be all the information and the small models. If you want to check out all the images that you generate, click on this output and then text image images here you'll see your output, for example, this one. Let's go back to our web interface here. Here, because we added the control net. You'll see the control net here. You can expand, you can upload your image. I will show you a little bit later all those settings and how you can adjust them to great images. But before I explain all the parameters here, I want to show you how to set up automatic 11 11 with a different cloud server called run diffusion. That's because some people may find run diffusion a bit easier to use, especially when it comes to using and installing your own models. In the next video, we'll cover run diffusion and then we will talk about different settings and how to create great images. 91. AUTOMATIC1111 RunDiffusion Setup: Another cloud server that you can use to run Automatic 11. 11 is Run Diffusion. Let's click here. Here is their website. Run Diffusion.com here, you just can click this. Get started, you'll need to sign up here. I already have an account for them here, it just locked me in. Okay, here you can. Here you can choose what server to use to run Automatic 11, 11. They have different plans. Here is the small server and it costs $0.50 per hour. It's 3 seconds image generation, it has stable diffusion. 1.5 and 2.1 300 gigabytes models loaded. Then latest automatic 11, 11 and Deform, And then two minute wood or launch time. Then we have a more powerful server that costs $0.99 per hour with 2.2 seconds image generation. The highest one is the 2.5 dollar per hour with 1.6 second image generation. Here are a few more servers, but for automatic, I would recommend using any one of these. Also, when you just sign up, you get 30 minutes for free to try it out. For example, to use the 30 minutes, I would recommend using the cheapest one you get accustomed to. All the features here you can see you can use the remaining balance. And here is 30 minutes. If you choose, let's say more expensive one, then it's only going to be 6 minutes because it's based on the credit here. They give you basically $0.25 for free in order to benefit from all the features you would need to sign up for the Creators Club. Here you get 30% of the large hardware, the 2.5 dollar, 1/hour Then you have 100 gigabytes of private storage, $6 of starting balance. But I guess the most critical element here is that with Creators Club, you can upload and merge models. So basically you can upload your custom model here and use it to generate images without the Creators Club. That's not available. If you want to sign up to Creators Club, you click the sign up here. So here you enter your payment information. If you want to use my promotion code, it's Caterina 15 and it's going to give you a 15% discount on your first month. Okay, let's try it out. So I'll go back here. I've already signed up to Creators Club, and this is what you'll see. You'll still need to choose which server you want to use. You get the discount with the Creators Club. Let's select this one. Okay, I have, my balance is $5.35 here. I can choose for how long I want the session running. Let's choose 1 hour here, then here it's automatically selected that. It will play down to notify when the session is ready. Okay, let's launch it. So it's going to take some time to initialize and launch and you'll be notified by the sound the first time you launch Automatic 11, 11. Here you'll have the username and password filled up. Here I have the S D user. Let's click Log in here. And here are Automatic 11 11 fowlers. For example, here you can add your model and I will show you how. Here we see Automatic 11, 11 interface. At the top here you'll find a timer. I chose 1 hour. It counts down that 1 hour. Let's say you change your mind and you want to make the session shorter. You can always click this Stop button and it's going to shut down the session. If you want to extend the time, you can click this extend button here on the top left. We can choose what model we want to use to generate images. And Here are a bunch of models here. There are base models like stable diffusion of 1.5 stable diffusion 2.1 But there are also more fine tuned models such as Dream Shaper. And here are also different versions like Dream Shaper 76 and so on. You can choose your favorite model here, for example, let's, let's use just the basic 20022. And let's generate something with this model here. Let's generate our Landscape prompt. And here I'll just click January again. We have default settings here. If I want more than one image, then I can increase my badge count. So let's put four. It's going to be pretty quick because we've selected the highest server. Okay, great. So here are our results. All the output images you'll find in the automatic 11, 11 images, just click here. Every time you start a session, you'll have a new folder for each day. Here is the 21st, let's use this one and here are my castle images. Okay. Now let me show you how you can upload your custom model here. Remember from the previous module, we've trained and downloaded our custom model here with the KPT extension here. You can just drag it to automatic here and it's going to load. Now I will also show you a different way you can upload it a little bit faster. Because here it's going to take maybe an hour to upload. But let's say you've already uploaded your model to Google Drive. Then from Google Drive you can upload your model here in a few minutes. Let me go to my Google Drive here. I've just dragged my model to my Google Drive. Let me find where it is. Okay, here is my model, SKS woman. I've renamed it to make it more clear here, all I need to do is go to the small actions here. I can go and share. Share. Okay here, make sure that the axis is anyone with the link, then copy the link here. Let's go back to Run diffusion here. To upload your model from Google Drive, just click these three dots here and click to shell here. You can put down and then Space Space. Paste your URL link, okay, here. Then click Enter. Now it's going to upload it, which will be a way faster than just dragging and dropping the file from computer. To learn more about different commands that you can use here you can go to run diffusion and let's say documentation. Here they have some guides and questions and answers. They also have a disc account. You can ask your question there to see if you have the model. You can reduce this window here to shell. Here we have our model that we got from Google Drive and as you can see, it was way faster here. Our model from computer is just uploading. It's not even halfway there. Now very important we need to move our model to this models folder. Let's drag it and drop here. Okay, now it's going to be in models here. You will need to move your model into the correct folder. Here we have version 1.5 it was stable diffusion 1.5 It should be in the version one. If you have models that were trained with stable diffusion 2.1 then they will be going into this version two folder. Let's place my model here in the version one. Okay, here I got my model here. Now all we need to do is just reload this website. Reload. And also let's refresh here. Okay, now we should see our model in the list here because the bunch of models, the easiest way is just to put the name of the model. Let's put cat here is the model. Model a woman. Okay, here it is, now loading. Now I can generate some prompt with my model here. Let's try again using the Astra prompt. Fashion photography portrait. I'll go back here and paste the prompt. I'll paste the negative prompt. Okay, let's generate. But here I will increase the badge count to four. Generate. Here we've got some interesting results with what looks like my face here. I like the dress from flowers here, but there are some artifacts in the face. And D as well. Okay. If you notice as when they have the prompt here, they also specify the parameters that used. For example, here you can see that what schedule they used, they used oil, the size steps, and so on. We can also try to use that here. Let's put 30 steps. Let's choose oil as selected here. Then let's they also use the face. Correct. Let's use that one here. Restore faces. Sure. And also yeah, let's try that. Let's check this out here. I think this one is pretty good. We don't have the lines on the face here. Let's see the other one lines again, but overall this is a bit better. I like this one. This is how you can use your custom model with Automatic 11, 11. And here you can generate whatever you want. Try different settings. In the next video, we will talk about those different settings and also the control net. Here, by the way, here are some useful commands that you can use with run diffusion. We've already tried the down that allows us to upload our model from Google Drive and that's way faster than uploading from a computer. Then with these commands, you can also upload model from any other link. For example, you find a cool model on some website, for example CDAI, that has bunch of cool models. When you want to use a specific model such as this one, you can just go and double click on this download. And here you can see the scopy link address. Then let's go to run diffusion here. Again, total the shell here. Let's put two space x eight. Then let's base our U R L here. Here is it. And it's going to download. The download is complete. Let's see, Let's reload. Okay, here is at the model, it's around 2 gigabytes. Now let's move it to models. Here, let's check out which is it is one or version two. It's version 1.5 Let's move it to version one here. I can also rename it if I want to. Here I can go and edit it. For example, I can put my model like this, rename. Okay, so here we have my fun model. Now let's refresh. Okay, let's check it out again, I'll use here, we have my fun model here, for example. I can use a girl here. Let's put for work. Let's, I've generated an image with this model that I just took from Vit Models like this one would be in run diffusion so you don't have to upload them. For example, if I just try to look for it. I B, as you can see it's already here and you didn't have to upload it. But I just wanted to show you how you can upload a model from a website like Bit AI here. We can also choose in painting version and so on. As you can see, there are a bunch of different models to choose from. 92. AUTOMATIC1111 Basic Parameters: In this video, I'll be showing you different settings and parameters that you can change in automatic 11, 11 to achieve better results with your image. Throughout the course, we've been talking a lot about stable diffusion. How to write prompts for stable diffusion. We've also covered different parameters here. It's going to be a great summary of the course because automatic 11 11 here, we'll find all those settings. It's going to be a good revision of all those terms. So first you want to start with a good model. It can be a basic stable diffusion model, like version 1.5 or version 2.1 Or it can be a more fine tuned model, like a dream shaper or I chose this ICBI L model, I quite like it. It's great for Porter realistic images. Then what's very important is to have a good prompt. We've covered that in Pro writing, where I explained some tips on how to write good problems. Let's quickly revise that. Here I have photograph, mid shot photograph of beautiful Brazilian woman. Here is our subject. Then we have a subject description in a bush jacket, extremely detailed ice. Then we have our background in a wild jungle through the foliage. Then I put in the style of dark brown, iconic sharp focus. Now we have lots of stylizerstendreyle of Jessica Dwon and Greg Rod Koski. These are our artists. At the end, I've also added 16 K HTR. We can move that before the style, so it's more consistent. Okay, here is our prompt. Now we need a good negative prompt. And for negative prompt we can use something like this. You'll find this prompt in the prompt presentation, cop head, bad framing and so on. If you want to highlight anything in your prompt, you can put that in parentheses. For example, Stop, that's important. And you can put 12 or more parentheses for now. I think that's good. Another thing here in Automatic 11, 11, there are styles you can actually save. For example, if you want to save this negative prompt the next time, you don't have to paste it here. All you need to do, okay, let's delete this first. Here, you just click Save button. And then you choose how do you want to save it? In my case, let's put just negative, Click okay here. Now you'll find that in the styles, you can just click here, choose the negative prompt. And if you write any prompt here, for example, our prompt here, the negative prompt will automatically be applied. Let's try it out. Let's generate. Okay, here. If we go down here, we can see our prompt and we also see the negative prompt. As you can see, even though we didn't put anything because we chose it to include in the styles, now it's automatically was applied. Similarly, you can do that for the prompt. Let's say you want to save stylizerst'st in the style of dark brown. Again, you can save that as a style. Let's save it. And let's put a photograph style. And let's click okay. Now we can choose our photograph style. We can choose negative prompt. Here, we just need to put our subject in the description of the subject. Here I'll put mid shot photograph of a beautiful Brazilian woman in a bash jacket. Extremely detailed eyes in a wild jungle. Using those styles, it's going to automatically apply our stylizers as well as the negative prompt. Let's see how it works again. Let's generate again. If we go down here, as you can see, our stylizersre automatically applied the same as the negative prompt. This is how you can save different styles for the genres that you work with. And that's going to simplify the whole process. Okay, now I'm going to delete that. Well, actually let's keep that. Let's choose the negative prompt and our photograph style. Okay, perfect. So now here, let's go into the settings. Let's begin with the simpler ones. So we have this badge count and badge size. The badge size is a number of images generated at the same time in one badge. Badge count is the number of badges generated one after the other. The differences between those two is that for badge size it requires let's say badge size of four or six. Because they're generated at the same time, it requires a higher V realm. It's more GPU heavy than the badge count. In terms of generation time, they are pretty similar. Let's actually try it out, for example, badge count. Let's put two. Let's okay, so here we've got two batches with one image in one badge. Let's change that. Let's put badge size with two. Again, here we've got two images, but this time they are in the same batch. Now let's move to width and height. For width and height, I would recommend sticking to the native width and height as was used for training the model. For example, this one is using the stable diffusion 1.5 version and was trained with the resolution of 512 by 512. This is the native resolution. If you want to make the image with higher resolution, then you can change the model. That's let's say 768 by 768. Or let's say you want to generate like twice as big as this, then you can use something like higher fix. We'll also talk about that a bit later. The problem here, if you just choose higher resolution here, let's say than something 1,000 by 1,000 then you may get strange results like double heads and so on. For that reason I would not recommend doing that. But let's say if you want to have the image in a different aspect ratio, then you can also change that. You can use the aspect ratio calculator to help us. Let's say I want the ratio with 16.9 Here are the width and the height. Let's use that. Okay, let's try that. Okay, as you can see here, we get this undesired result with the second replica, this woman here. Let's actually try the same ratio, but make it scale down by the factor of two. Let's put 640 by 360. That's going to give us the same ratio. As you can see, it gave us the same ratio, but now we do not have that undesired result. You may have to play around with height if you want something else but the square. 93. AUTOMATIC1111 Parameters - Sampling Steps, Sampler, Seed & CFG Scale: Let's move on to sampling steps. Steps are the number of noising iterations in the generation process. As you remember, stable diffusion starts from random noise. With every step, a new information is added eventually to get to the clear image based on your prompt. Let me show you some examples. Here are sampling steps. As you can see, when we have only one, it's a very blurry image. But as we increase the sampling steps, we get more and more information here. This is at five steps, it's still a bit blurry, but at least we see the eyes. The nose of husky dog and then at ten it's now a clear image. And then the more we increase the steps, the more details we get. But there is not much difference as between, for example, step 1.5 or 5.10 Here, it's very small difference between steps 20.30 or 20.50 Maybe there are slight changes but barely visible. I would say around 20.30 steps is a good choice, but it really depends on your image. If you want to make something more abstract, maybe you can choose five steps if this is the image that you try to achieve. Okay, let's get back to our run diffusion here. Let's just try with this image. Let's put, let's say ten. Let's, let's see here. We've got some details, especially on the jacket here. But overall, especially like the hair face is quite blurry. Let's pump it up. Let's say 40 steps. It's going to take longer time to generate, but let's try it out. Okay, here, let's check it out. You can see that the hairs are now way more detailed. The background as well has more details. This is how you can adjust the sampling steps, even though there are quite a lot of details. But there's definitely a problem with the face here. Here we have a button called to restore faces, and we can use that to help restore the face here. I'll copy the seat so we'll get the same image. So seed. Okay, let's generate again. Now you can see that the phase is way better. And that's what restore phase button here, what it does. Okay, Now let's move on to sampling method. Okay, here for sampling steps, let's make it 30 because I think that's a good average. I will also disable the restored phases because it also takes longer time to generate images. However, let's say if you liked a certain image in your badge, you can always restore base on that specific image. Okay, for sampling method, sampling refers to the denoising process in stable diffusion. And there are different sampling methods that can produce a bit different results. Some they will converge to the same image, but others will produce slight differences. By default, oiler works pretty good and it's also a quicker sampling method. What I mean by quicker is that it's going to give you good result with a number of sampling steps at around ten or 15 sampling steps. It's going to give you a clear image, maybe not with too much details, but it's going to be a nice image. Actually, let me show you all those different sampling methods. Okay, here on the right we have different samplers like oiler and you can see that at around step 16 we already have a good image, and another good one is D D IM then we also have A new samplers like P M plus plus two Mars. These are newer samplers and we'll also produce fast and good results. Maybe let's choose that 12. Here we go. Okay, let's try that one. Here is the result with the PM two, may I'm going to switch that back to oil. Now let's talk about the CFG scale. Cfg scale controls how much image generation process follows the text prompt On different platforms you can find CFG scale could be referred as Prompt Guidance, guidance scale, or even in some, I've seen that as prompt weight. These refer to the same setting here, basically, how much generation process follows the text Prompt. Where lower values mean that the result is going to be more creative. Higher values will be better aligned with the prompt, but may create a higher risks of artifacts. And I'm going to also show you the default is around seven and the recommended range is around four to ten, depending on your image. Let's see. Okay, here we have two different images, one with the scheme here, and this is with the CFG one. You can see that usually a lower CFG will also have less details and will also be more pale. Whereas with a higher prompt guidance, you'll find images that have more contrast. Overall, they are more saturated, so lots of colors. Also, if you take a look at this prompt guidance of 15, you would notice some artifacts starting to show up. For example, the fur starts to feel unnatural and more sharp, whereas before that, maybe at 75, it's way nicer. Let's try it with our image here. Let's maybe CG scale two, so you can see the contrast G two. Okay, let's generate. Okay? You can see here that the colors are more dull. Let's try with the higher scale here. Let's maybe put around 11. Let's use the same seed. Okay? Okay. Do you see the difference here? Colors are way more bright, 6-8 It's going to be a good balance in colors quality and you'll get image that matches your prompt. Let's put eight here. Okay, Now let's talk about the seed. The seed is initial input that guides the creation of the image. The same prompt and parameters will produce the same image with minor variations here, by default, you'll have a random seed every time image will be generated with a different seed. But let's say you liked a certain image, then you can use the seed. Let's say I like this image here. I can click this green recycle button. It's going to copy the seed number of that image. Another way to see the seed is to go below here. Here at the seed. You also can get the same number. Okay, let's say if I change some settings here, let's say I want to restore phase. I've copied the seed, now I'm expecting very similar image, but now it's going to have better quality phase. Okay, as you can see, almost the same image and minor changes in the face here. That's how you can use the seed. You can change some parameters. Photograph of beautiful prettily woman, let's put smiling. Hopefully that will only change her emotion. But we'll keep the composition the same here. I think it changed a little bit too much. We have new clothing details. It really depends how you modify the prompt and what effect it will have on the image. Here. Sometimes it will just have minor changes. But with the same set, you will get a completely different image. 94. AUTOMATIC1111 ControlNet: Now, what happens if you want to generate images in higher resolution? This one is 512 by 512, which is a pretty low resolution. What if you want to generate twice as high? Remember when we've talked about width and height, I told you that it's not a good idea to just modify the width to twice as much because it's going to produce strange results. And you'll get double head, double bodies and so on. For that there is this button called Hire Fix. If we click on it here, let's disable the restore phases. You can see that you can resize an image 512-1024 This is an upscale by a factor of two. Let's say you want it even higher, then you can change that to three. It's going to be 1,536 5,536 Important things to change is this upscale by factor, you can do 23 or four. The higher the factor, the longer it's going to take to generate the image. Another thing that we can change here is the higher steps, we now it's at zero. For higher steps, you can set it to zero for pure image upscaling. If you don't want any changes to the image you want it as, then you can set it to zero. But usually it's nice to set it maybe two as half as many steps, your sampling steps. Here we have 30 sampling steps. Let's put 15. Another parameter that you want to change is the denoising straight. The denoising strength controls how much the image will change. Near zero, no detail will be added to the image. Near one, it will completely change the image. I would recommend to use around 0.2 and 0.7 here. The lower it is, the more closely it's going to be to the image. Let's put your 0.2 for example. Another thing you can change is this up scalar. There are different up scalers you can use, but for now let's just use the default latent one here. If you have a random, it's going to generate a completely different image. First, it's going to generate image with 512 by 512. Then it's going to scale it by your factor here. In my case it's going to be three x. Okay, let me use a random seed here and let's generate. You will not see the 512512 image, you will directly see the higher resolution image. It's going to take a little bit of time. It took way longer to generate just this one image here. I even reduced my app scale factor because it was just taking so much time here. I changed it to two. Here is the result. The quality is way better here, but it just took so much time. What I would recommend is not using the higher fix and just generating a few images. Let's put like four images then choosing your favorite images from one of them. And I'll show you what we can do next. Let's see these ones. I think this one is pretty fun one. Let's use this one. Okay? The way you can improve the quality of this image is move it to image, to image generation. Let's send it to image to image. Okay, now we're in this image editor. I see how change the tab to image to image. Okay, here we have a few more settings. Okay, here you can choose to restore the face also. Let's resize it by here. If you want to set the width and the height, you can do that or you can resize by a factor here, it's easy for me to do the resize by a factor here. I will change it to two. Then again, you can change the prompt guidance scale. But the most important parameter that you'll be adjusting here is the denoising strength. Again, denoising strength affects how similar it will be to your image here. If we want to make it very similar, then we can set it to 0.20 0.2 here. If we want, let's say more details or changes, then we can increase the de, noising strength. And let me show you what will be the difference. Let's first create with 0.2 here. Next time I'll change it to 0.7 Okay, let's generate. Here is the result. Okay, Now let's change it to zero point. Let's actually put 0.85 We see the difference here. You can see that there are so many changes to the original image. Now her hair is covered by this hootie here. We also have a slightly different background. She's smiling way more here on this, her eyes are squinting. Quite a big difference. That's how you can use the Ois strength to choose how much you want the upscaled version to resemble your original image. 95. AUTOMATIC1111 Hires Fix and Image to Image: Now let me show you how you can use Control Net here. You can use it with image to image generation, but for simplicity, let's use text to image. Okay, Here if you go down here, we have this extension already installed. So we have this Control Net version 1.1 okay? And we just can open it to remind you we've covered control net in Leonato, More Agile. Control Net allows a reference image such as this one to influence specific attributes of the generated image. There are different models. The more popular models can, which takes edges of this reference image, which makes a depth map from this reference image, and then pose to image. Pose to image can be just the pose of this person here. Or it can be together with a facial expression. You can read more about control net here in this article. But let's get back to run diffusion. Okay, here we've opened control extension here. Just applaud your image, your reference image. For example, let's use the same one as we've used in Leonardo module. This one, Okay? Here, very important, Make sure that it's enabled. Because if it disabled, if there is no tick here, then it's not going to use the control net. So make sure it's enabled. Okay. Now here, let's choose the model. There are different models, and there are many of them just for let's say depth to image. There is four of them. For open pose, there are even more. There is open pose full, open open pose face, the open pose full is a pose with the face. The open pose is just the pose without the face. Yeah, let's use just the pose here. We chose our preprocessor open pose automatically. We'll choose the model for us here, which is nice. Then to see how it will process this image, you can click Allow, Preview, then click on this explosion mog. Now it's going to process the image and we'll have the control net map. This is what it will feed to our image generation. Let's say if you choose a different processor, let's use the open pose full again. Let's click this explosion here. You can see that now it also includes the phase. Okay, I have some problem with scrolling here. Scroll manually. Okay. Here, open pose. Let's use the open pose. The next important parameter here is this control weight. You can adjust control weight to increase or decrease the influence of control net on the generic grid image. Let's say, let's actually try something first. Let's use open post Okay, process. Now here let's click Generate. One more important thing, if you use ad blockers or if you use brave browser. Control net may not work with run diffusion. If you have a problem, control net is not working, try to switch to a different browser. Okay, here we've got our four images. And you can see all the images, they have similar pose to our reference image here. Now let's reduce the control weight and see what's going to happen here. Control weight is one. Let's make it, let's say 0.2 That's going to reduce the control net influence on the generated image. Okay, let's see the other images, the four images. As you can see, it still uses similar pose. But right now the poses are a little bit more diverse. That's because we've decreased the influence of control net. Here, with open pose, the control weight is not quite apparent. But if we use Y, for example, Y is the edge to image, it's going to use all the edges here. For example one. Let's generate here, you can see that mimics our reference image. The here, the color of the dress is different, but everything else looks very similar. Now let's reduce the control weight. Let's make it 0.10 0.1 Let's generate again here, even though we've used the same Can model, basically we've used the same settings except for control weight. Now the images have way less influence from the control net map then previous badge. Here you can see that some images still have let's say similar pose but this one is straight looking as you can see. Way less effect of control net on these images with the can. I think it's pretty apparent here. This is how you can use control net to influence your images, which is a very useful tool. 96. AUTOMATIC1111 Inpainting Extras: Now I want to show you how you can use in painting in Automatic 11, 11. For that, I will disable the control net here. Okay? It's not going to be affected by control net here. I will go to image, to image. Actually, before that, let's generate a full by short photograph. I can show you fully. Short photograph, that's good. Okay, even though we put here full body shot, we're still getting a meat shot here. I will just add the trousers. Okay, let's beautiful Brazilian woman. Okay, let's take, smiling away in brush jacket then and safari trousers. Poor eyes. Let's just take that out in a wild jungle. Okay, that's good. Let's try that. Okay, let's check this out. Okay, so this is more of what I was looking for, okay? I like the second one, okay? See how the phase is screwed up in all of them. And that's because there's just very small area for AI to generate good quality face here. For that, you can go and try to upscale. Another method is you can send the image to paint. Let me show you send to paint. Actually, let's first choose the image here. Let's use this one. Okay? First of all, let's increase the resolution of the image a little bit. Send image to image. Okay? Now here I will first increase, I will resize it by two here, denoising strength, Let's make it 0.4 0.5 Yeah, I think that's 0.45 I think that's good. Okay, now let's do that. Okay, so here we've got the image that's way better. We've got more details on her shirt, her pants. Okay. Now, but as you can see, face is still not the best quality for that. Let's send this image to paint. Okay, Now it's in the paint tab here. We can zoom in and use this arrays tool to erase the face. Okay, Here if you want to change the size of the brush here, you can change it. But right now I think the swing would be good. Okay, now I've erased the face here. If we scroll down, here are the settings. The most important parameters here are masked content and paint area. The first masked content as if you want to keep the original quantent underneath your erased heart. Let's say if you want to keep a very similar phase here, then you should choose original. If you want to generate a completely different phase, then you can use either latent noise or later nothing. Then we have paint area here. You can select whole picture or only masked the difference will be in the resolution of generated image. Let's say if you choose only mask here, the resolution will be way higher than the whole picture. Because here we've selected just this small area here. And it's going to be using the same resolution as this whole image to generate just this area. Which will give us way better results. Especially if you work with faces that's ideal to choose. Only masked, then you can also adjust the denoising strength, and that's going to influence how close it will be to the original content. Let's say I want only minor adjustment to the phase here. Then I can move my denoising strength lower, so make it 0.3 or two. But if I want a bit different phase, then I can move it higher and make it 0.9 If you make it the highest of one, then the phase will not resemble anything like the original content. So for noising strength, let's put 0.5 because I want to see some differences. Now let's change the prompt here. We want to generate images of the face here I will full by D shot here. I'll just remove that photograph of beautiful Brazilian woman Here I can put detailed is something that refers to her face. Okay, I have my negative prompt and photographers. Great, let's generate. Okay, now we've got quite different phase but now it's way higher quality. In my experience with, in painting, you would get way better faces than if you just click restore phase here. Restore phase, you can do that, but from my experience in painting will give you better results. Now let's try a few more things with, in painting. Now I like this image and I will move that to paint again. Okay, here I want to show you how you can remove stuff from the image. Let's say I don't like the earrings here. I just want to remove them again, Just paint over the element that you don't like here. Okay. Now let's adjust the settings here. Again, I want to keep the original even though I don't want to see the earrings, but because here I have the hair around, so we want to have the same hair. I'll keep the original down here. I will just move the noising strength to a higher value, maybe 0.9 We won't get the earrings up here. In the prompt, I will also put photograph of a beautiful Brazilian woman here, hair, and I don't put any earrings. Let's try to generate this and see if it's going to work. I also increase the batch number so we have more images to choose from. Here, I'll choose form. Here are some results. Instead of the golden earrings, I tried to add the leaves, even though we didn't say anything here in our prompt. But let's see, other ones again, some leaves. Okay, The best one is the last one. But, but what it added here, I don't think it quite matches her hair style for that, if you want to remove something from the image. There are actually better models that you can use for, in painting. If you use front diffusion, you can just go and search in painting. Here are all the different models that have been trained to do in painting, such the basic one, we have the stable diffusion version 1.5 in painting. If we run it again, the results usually from what I found, we better if you're trying to remove certain elements from your image. If you are changing a phase, then again, from my experience, I usually like using the models that are best for pass. Not necessarily in painting, but just either Photos models, the ICB, INL or Dream Shaper, which give a really good phase results. Right. Now we select the painting. Here we have the 1.5 grade. Again, we have everything deleted here. Let's check the settings again. We have original At the bottom here we have the noising strength, 0.9 which is high. We don't want to see any of the golden earrings. Let's try that. By the way, I also changed my prompt a little bit. I put coiled hair. Okay, let's see the results. Okay, On the first one, I see that it perfectly removed the area. Let's see the other ones. That's good. I think the second one is the best, the shape of the ear, especially in the neck. Let's choose the second one. Once you've got the image that you like, you can actually upscale it even more For that, just click Send to Extras. So let me show you what it's going to do. It's going to open this extra tab and we'll upload our image here. Here, basically we're not changing the image anymore, we just want to upscale it as here. All you need to adjust is this re size factor. Here I can upscale four X. Also very important. Choose your upscale. Here in run diffusion, we have a bunch of up scalers. The default one you get with automatic 11, 11 is the RS gun X plus, let's use that here again. Here we have a factor four. If you prefer writing the width and height yourself, then you can use that, but for me it's easier just to choose a factor. Okay, Now once you adjust the settings here, you can click Generate. Here is a final result which you can save. Save image as save grade. Now the image is 4,680 4,608 Let's open it here. And now you see that the quality is way better. Maybe I would work a little bit more on her hair, but overall, everything else looks really good. Another thing is that each up scalar is a bit different. You may want to also play around with different up scalers. For example, if you work with anime, there are specific up scalers for anime that work better for anime styles. I've summarized all the information that I've told you in the Powerpoint so you can use that for reference. I also listed useful links that you check out. Here are some really good guides for Automatic 11, 11, Especially because Automatic 11 11 is a pretty difficult app. So it's going to take some time to learn and master it, try it out, play with different parameters and settings, and challenge yourself to generate some cool images with it. In the next module, I'm going to be using different AI tools that we've covered in the scores to show you my workflow when I'm tasked to generate specific images. 97. Create a Comic Book Version of Yourself in Astria.ai & AUTOMATIC1111: Hello? Hello. In this module I'm going to show you my workflow on specific project. So let's get to it. Okay, so the first one is pretty easy, AI selfie, create a comic book version of yourself. So for this project I will be using Automatic 11, 11 and the model that I trained in, Astaire. Let's go to run diffusion here. Okay, here I will find my model that I've trained in, Astin, I have uploaded to run diffusion and it's model SKS woman version 1.5 staple diffusion. Here when I renamed my model, I usually like to add the instance prompt which is in the SKS woman. So I don't forget, I need to make sure that I add the SKS woman. I have already prepared a prompt for comic book images and I used Astra as the inspiration here. They have a really nice prompt, 23 with a comic portrait of cyberpunk SKS woman, and the images are really nice. I've used some of the prompt and then also added more of anime. Okay, here let me paste the prompt. Here I have a comic bulk of a cyberpunk superhero SS, Woman with big and cute eyes, curly hair, Comic book phase, fine details, night setting, very anima style anima style. Or here is repetition. Manga style hand drawing, cinematic sharp focus illustration. Big depth of field masterpiece, concept art. So lots of stylizers trending on art station. And then I specified the style and the artist, and then I gave the weight of 1.2 See how my SKS woman I used to parentheses. This has the highest weight here. Okay. Now, let's also add the negative prong and I've also prepared it here. I just have lots of words like to form blurry and so on. Okay. Before we change any settings, let's try it out. Okay. Here for badge count or batch size, I only have one image. Let's make it four images, Let's check this one out. Okay, it captured some of my features. I'm not sure what it is on the forehead here. But let's try one more time. Let's see. Okay, that doesn't have my facial features whatsoever. Let's see the other ones, he is not too great. The phase is a bit messed up. See, let's try to do restore phase with that. The restore phase makes the face less anemic comic book style. I don't think I want to use that parameter, I'm going to disable it. I'll just try generating a few more batches. Okay, let's see. This one is way better, but again, something on her forehead. Let's see, the other ones definitely better in terms of facial features. Okay, not too bad. I'm wondering if we should change anything about the prompt. Maybe in the negative prompt, I will add photo realistic. Maybe for Comic Book, I would also add more weight. Okay, Comic Book as woman would pick. Yes, I think that's a pretty good, let's try that. Okay, this time, let's check out. I like the result here. It's comic book style, yet it captures my facial features. Let's see the other ones. Yeah, not bad here. Okay, this is to enemy. And this one is also pretty nice. Okay, now that I know that the prompt works, I want to generate my images in a specific pose. For that, I found a cool image of Tony Stark, and I want to recreate that pose for my character here. Okay, and for that I'm going to be using control net. So I'm going to open it and I'm going to upload the image. Here is the Iron Man create here, make sure it's enabled. Then here I'm going to use the depth model. Let's preview it here. I'm going to click on this explosive G. This is the depth map and control weight is one. Let's try it with that. Okay, here, let's generate again four images. Use that. Okay, let's see. I like this one. This one seems too realistic but is also not bad. Okay, actually I want to use the seed for this image. You get this image again for seed. If you want to copy the seed, you can click on this as I call it, Recycle button. Here, we've copied the seed of this image, full control net. I want to change the control weight because I think the pose is forced, so I'm going to make the control weight smaller. So here we have one. I'll make it 0.7 okay? And let's generate, okay? Okay. Let's see. Okay, So the first one, okay? I actually really like this one. I'm probably going to be using it. Let's see the other ones, this one is not too bad, okay? I like the first one, but there are a few things that I want to work with. For example, I don't like this red button here, so I'm going to work a little bit more on the suit. Okay, so here I'm going to send my, the first image to paint. Okay, let's zoom in. Okay, first I'm going to work with the phase. I'm actually going to replace all of this here. Now for the settings, let's choose original for mask content here, we'll choose Only Mask to increase the resolution of the mask area. For denoising, let's make it, right now it's 0.2 Yeah, that's good. It's going to change a little bit, but hopefully that's going to increase the quality of the face here. Let's remove the night setting because it's not relevant to the face. I think that's pretty good. Also, Curly here, comic book phase. Okay, let's generate. Okay, so let's check out the results. I think here I've got too many wrinkles. Let's see the other ones, okay, I think the best one is this one. But I still would want to correct some things here. Again, let's send it to paint. Okay, Create. And now can remove small details here. And here I'll just put me and the skin. And let's check our settings. We have original only masked. Then denoising strength 0.2 Let's make it the highest because I don't want the wrinkles 0.8 That's going to change that. Let's generate. Okay, here we've got some strange results. I'm just going to go and make the denoising strength lower. As you can see, you just have to play around with the denoising strength. Okay, let's make it 0.5 Let's generate, okay, I'm still getting strange results now. I'm just going to choose the in paint model in painting. Okay, let's use the stable diffusion 1.5 in painting. Let's take a look here. The results are way better. And this is what I was looking for here. We just changed the model to stable diffusion in painting. Model that was trained specifically for in painting. And we kept all the other settings the same and see how much better it is. Okay, let me show you the settings for settings. I had mask content original, only masked in painted area. And for denoising I had 0.5 But here with this model, the results are way better. Okay, now let me choose the image. Let's see the other ones. I think the third one is the best. I will send it to paint. Now here I can also work on some details. For example, the shot, for example. I can remove this orange pattern here that I don't like at the top here. I'll just put Superhero Shot. Okay, for the noising, let's make it a little bit higher, 0.7 That should be good. Let's generate. Okay, now we try to incorporate the Super Man logo on the suit. I don't want that actually. Instead of superhero, I'm just going to put blue suit. Okay, this time it's a bit better. Let's see the other ones. This one is pretty good. I like the second one, but I think we can work a little bit more with that. I'm going to send it to paint and work a little bit more and we'll show you the final result after a few painting. Here is the result. Now I'm going to send it to extras and we'll upscale the image. I, let's use the anime here. Let's upscale for X. I think that's good. Let's generate, let's check out the upscaled image. Okay, I think it turned out pretty good. I'm pretty satisfied with this image for our first project here. 98. Create a Book Cover in Midjourney & AUTOMATIC1111: In my second project, I'm tasked to design a book cover for a new edition of Alison Wonderland adapted for children ages three to five. And here are some specifications, incorporate key elements from the book. It should be illustration with bright, vivid colors, with a resolution with at least 1920 by thousand 80, which has aspect ratio of 16 to nine. Let's do that. For this, I will start with N in my Discord account here. I'll just imagine, and I'll, because it's a famous book, or us should know about Alice in Wonderland. Then here, book cover. Let's see what it gives us. Okay, Here we've got our images. But I think for children's book, this is too much. First I'm going to go and look for some inspirational images on the internet. Some children's book covers. Okay, here I put Alice in Wonderland book cover. And I'm going to go and choose the one that I like the most here. I like this one. I think it's pretty simple. If we zoom in, it's pretty simple. And I like the cal palette here. Okay, I will save the image. Now I want to use this image for mid journey. The only problem is that it has a lot of text. I don't want I to be influenced by the text. I don't want it to generate more text. For that, I'm going to go to clip drop here. In the tools we can choose, either clean up or the text remover. Let's start with the text remover. Here is the image. Okay, the text remover was pretty bad. Actually, I'm going to cancel it. Go back to clip drop and choose the clean up tool. And then again choose the image here. And then I'll just erase all the part that I don't want to see. Okay, now let's clean this. This is way better than the remove text one here. I will download it. Great. Now I want to use it as the inspiration, but first I want to understand what style is a so I can give better description to my journey. I'm going to go and use the describe Commando Ter. Okay, here we have Alison Wonderland in the style of whimsical children's book illustrator. Light orange and so on. Fairy tale. Right now I'm looking for the style storybook like soft edge. Okay, soft edged. Then we have watercolor illustration from here, for example. I can use soft edge and watercolor illustration. Now I want to use that as a reference for my prompt. I will use the image address here. Copy image address here again. I'll click, imagine I'll paste the link here. Now I will write my prompt. Let's put book cover, Alice in Wonderland. Now I'll describe what I want to see in the scene. I want a small girl in a blue dress in sun lit with flowers following a rabbit. Let's try that. Actually, I forgot to add the style here. I'm going to do another one. The same prompt, but just add more details here. I'll put soft edged and then water color illustration. Let's check this out. Okay, I think the fourth one is pretty cute, but there's definitely problem with the rabbit. But it's okay, the rabbit is not that important. I can always paint and change the rabbit. Okay. But for now, let's scale or let's generate more versions of the number three here. Let's see the variations. Okay, I think this is pretty cute. I like the first one. I can upscale it upscale. The first one here is our upscaled version, which is pretty good. Let's see the other ones. This one was with the soft edge watercolor illustration. I think here we have too much of the watercolor aesthetic, although maybe the fourth one is not too bad. Okay, let me show you my settings here. If I go to settings, I'm generating images with the latest model, 5.2 and I'm using the stylized medium. Maybe I can make it a little bit lower so we have more simpler images. Let me try that. We can, I'll copy the prompt here. I like the images without the watercolor illustration. I think they came out a bit better. We'll continue using the more simpler prompt here. I'll put imagine, I want to change my stylized parameter, so I'll put stylize 100. Also, we were tasked to create image in 16 to nine. Aspicratioiill. Also add that. Aspicratio 16 to nine. We need to space here, great as generating a bunch of images with the same prompt. Here is the result that I liked in particular, number two here. For the prompt, I kept it the same, but I added the negative, no trees or tree. That's because I was getting just too many trees in the image. Now, even though there are still trees but way less dense, then I've upscale the image. Here is the result here. I think it's too zoomed in, it's best to zoom out the image I did, particular that I zoomed out 1.5 x here. The result which I really liked, the only problem here is that the resolution of the image is not great. It cannot be used anywhere. Also, I want to change the rabbit here because it doesn't match the Alice in Wonderland theme. Okay, the girl is great here reading a book by the tree. In order to upscale the image, there are different services we can use. For example, Big GPG. For Big GPG, we can upload the image, Let's choose four X art work and noise reduction. Let's move it to high. Okay, great, let's start here, Let's check out the result. Now the image has way higher resolution, but I think in terms of quality, it's not great. We have some problems with the hands here. I also want to remove the rabbit. Actually I'm going to use this image and put it in stable diffusion. Automatic 11, 11. Let's do that. Going to close this image. And I'm going to go to run diffusion here. I'll go to image, to image. I'm going to open this in paint and I'll upload my image here. Not the upscaled version, but the mid journey version. Here we have our image. I can move it. Okay, here, I want to change a few things. I think her face is not sharp. I'm going to change that. But before I do that, I want to change the model. I want to use a model that's great with faces, so I think Dream Shaper should be good for this kind of image. Let's try that. Dream Shaper. Dream Shaper seven, Okay, now it's loaded here. For the prompt, I'm going to use something similar to mid journey Alice in Wonderland. A small girl. Let's just keep a small girl here. I will raise her face. Okay, Now let's go to settings down here. Mask content, Original pad area A masked to increase the resolution. And then for denoising strength, I want to keep it as close to the original as possible. So let's maybe put 0.2 Okay, now let's generate. Let's check this out. Okay, not my favorite. Let's try one more time here. For the prompt, I will actually add the style by Excel Scheffler. Also for the negative prompt here I already have some negative prompts, I will use that. Okay, now let's increase the count for the batch count. We can generate more images. Batch count three, de noising strength, let's make it even less, it's more similar to the original image. One point, let's put 0.15 Okay, let's check them out. Okay, this time I think it's way better. I like the first one here. I'm going to use it to load to the paved. Okay. Now I'm going to go and remove the rabbit here. I'm going to use a bigger brush for this. I'm going to use a different model. I'm going to use the painting model, stable diffusion version 1.5 Okay, now here I just put grass field, it's a bit blurry. I'll also put blurry. Okay, here I don't need the negative prompt for the settings, I want them to be very different here. I can choose latent noise and then increase my de, noising strength. Or I can use original and literally like 0.90 87 or 98 to have very different image. Okay, let's try that. Okay, let's check out the result here. It added some flowers. This one is pretty good, okay? The first or third one did a pretty good job. Maybe I will use the first one here. I'll again set into paint. Now it would be nice to merge the edited part with the surroundings. We can easily do that. Again, I'll just my brush here. Let's change our de noising strength to zero point around 0.4 I think that's going to be good. Let's generate again, let's see. Okay, now it's a bit better here. Now as my last step, I also want to add a rabbit here behind the tree. I think that's going to create a nice composition. I'm going to use maybe the second image here and I'll set it to paint. Now in the spot I want to see a rabbit. For our settings, let's use original only masked here for denoising strength. Let's make it one because we want a completely different thing on that spot. Okay, and here I'll put a white rabbit watching from the tree. After generating a few more images, I had to change my prompt. I added the style by all, it was more in the same style. Then for the settings, I also changed a little bit here, I changed the Generis strength to 0.9 I like this one, but it Mrs. the ear. Let's see, the other ones, I think the bit is a bit small, this is just a mass. There are a few more things that we can in paint. Like our hands, the head bow here. I did it in the same way, was choosing an image, then sending it to paint, then removing the element that it didn't like, and then constantly improving the. Finally, I send the image to extras, and then I chose the four x ultra mix balance up scalar and I resize it by a factor of four. Here is the result. As you can see, I've merged the grass here. I changed the rabbit. I also improved the hands here, and I edited the head bow here. Then I've upscaled the whole image. I think we got pretty good results here that can be used for book cover. Also, remember at the Firefly module where I showed you that you can create fun letters. Let's try that Firefly, Adobe.com Here we can choose text Effects. If you want to use letters for a book cover, then you can also use this Firefly Adobe. For example. Something interesting with flowers, magical forest, for example. Here we can put Alice. Now we get interesting results that can be incorporated as part of the book. The only problem is let's download this. The problem here is that we will need to clean all of this up, which we can do with Clip Drop which allows us to clean up things. But another problem is right now you cannot use Adobe Firefly for commercial use. That's another thing. But otherwise, I think the concept is pretty fun that maybe in the future when they allow the commercial use, you can generate fine letters for book covers. Also, I want to show you from what we've started and our final result. Here is the mid journey image that we generated. And here is the final result, that's a way better quality that we were able to generate in automatic 11, 11. 99. Create a Logo in Leonardo, Midjourney, ClipDrop, Vectorizer and Firefly: For the third project, I'm tasked to design a logo for a newly opened space alone. Harmony and balance Spat, and here are specifications. Create elegant and captivating logo that embodies the space essence of tranquility and rejuvenation. The logo should have a modern dynamic and minimalistic style. Avoid excessive details focusing on creating a clean and memorable design. Okay, let's try it out here. I can use different platforms like Mid Journey Leica for Adobe Firefly. I would probably use it if it were available for commercial use because I think with the logos, we got really nice results with its AI generator. Let's actually start with the pond for this, I can go to Leonardo and go to AA Image generation from Generation. Here I just need some ideas for my logo and leonadogenerates nice ideas. Let's use it for example, Spa logo. I could also probably use GPT, but let's use Leonardo, since we've covered that in the course. Okay, A vibrant abstract logo design featuring a Trent keel pool of water surrounded by lash greenery. Oh interesting. Here we have the lotus flower that's very typical for spa salons. Modern logo design featuring a stylized spot building. I think that's too much stylized sun rising over Trunk Lake. This seems very interesting. Let's actually try that because Mid Journey is great with simple products. I'm going to use Mid Journey next. And he'll just put, imagine from over project we want the modern and minimalistic style. Let's add that minimalistic. Okay, okay, here we've got some logos, but I think it's beautiful but it has too much detail. Maybe I want to focus on a specific object. I'm going to go back to Ad and maybe generate more prompt Spa logo. Let's generate a few more ideas. In the meanwhile, I will also Google images. Many of the images here have the Lotus. Let's do something different but the Lotus or like the rocks here. Okay, actually let's go and just for mid journey, I will try very broad term. I'll just put Spa logo and I'll put simple Modern. I'll just put minimalistic and symbolic here. I will increase chaos because I want variation and chaos. Let's put ten. Okay, Let's see if Lado generated some interesting ideas, okay? Vibrant abstract logo design. Lotus flower, single lotus flower. Lotus flower. Okay. Everything is lotus flower. That's not fun. Okay? Okay, so here are some images I like the theme of the first and the third one here just looks strange like a mask, nose, mouth. This one is the lotus and it's gold. That's like a luxury spot. Not bad. But I want to make something different. Here in the images, I found that in one of the logos, they use candles. Let's try to use candles in the logo. Okay, I'm going to go back to Journey. Imagine and again, spl, let's portraying candles minimalistic, symbolic. Let's put harmony. Let's put chaos. Actually, I think let's increase the chaos a little bit. Maybe 20 in the meanwhile, I'm going to go back to Leonardo and put candles there. Maybe it's going to generate some cool problems with that. Spo candles featuring candles. More ideas. A single candles right by a ring of delicate petals. Bouquet flickering candle of vibrant flowers. That's nice. Surrounded by a, surrounded by a circle of vibrant blooms nested in a bed of vibrant flowers. Here I got the idea. Candles with leaves, flowers. We can try that starry night. No, I don't want, it just adds more details. Let's put minimalistic here. Let's see if that's going to help. Okay, so here we have minimalistic spy logo featuring a single candle. Okay, now this is more simple, modern spy logo with single candle surrounded by ring of liquin flames. Interesting. Surrounded by circle of soft light, glowing ember, circle of source. Okay, Circle of intricate pattern. That's interesting here. It's surrounded by circle of intricate designs. Basically the same thing. Actually, let's try the first one and the seventh one. We'll generate here. Generate. Okay, let's go back to mid journey. Okay, now we've got something interesting. This is basically what Leonardo was telling me. That the candle is surrounded by some flower. I like that, we can work with that here. If I like a specific image here, I can upscale it. Number three here, I will actually save it. I'm going to go and describe here. We've got some description here. We have in the style of light turquoise and light white. I like the turquoise color here, The energy filled illustrations. Indian traditional oil in a lotus flower logo. A candle with branches and leaves, a lotus flower logo with a candle burning inside it. I like a candle with branches and leaves. Here, I'm going to use that. I want something like this image here. It gives this luxury vibes. I'm going to also upscale it now. I actually want to blend those two together and see the result. I'm just going to click Blend here. I'll put first one and the second one here, dimensions we want the square, Let's use that. Even though it's still generating, I can see that we're not getting this black background, which I want. Let me describe this image as well. Describe, okay, so here we have sharp attention to detail. Okay, A logo. Okay, that's the name of it. It nicely took it from the image here. Gold in the style of serene peaceful ambience. That's very vague in the style of gold leaf. Okay, again, very vague. This is what I'm going to do. I'm going to take the description of this image here. I'm going to use the image address of this image. Okay, let's write a prompt. Imagine I will use the image address of this one, called the image address pasted here. Now I will further describe may prompt a candle. Actually, let's start with, it's a logo featuring a candle surrounded by petals, intricate patterns. This is what I took from Lead. Let's see actually what Leonato has generated. Okay, so these are like this intricate patterns, but it's too complicated and this one is too simple. It looks more like a photo than a logo. I'm going to stick with mere journey, but I like the ideas here and I will use them. Let's go here. Intricate patterns. Let's try that. Okay, this starts to look interesting. Maybe intricate patterns was too much because it adds too many details. I'll take bit of that. I'll change my pro. Okay here, I just can click control to get my prompt. I will change it here. I'll remove the intricate patterns because I don't want too much details. And also I will make stylize, let's put 50. It's more simple. Also I'm thinking if I want to make the image weight higher or lower here I got very similar images. Maybe I'll make the image weight a little bit lower. The default one is one. I'll put 0.9 I also want to add chaos. Let's put Chaos and let's put 15. I'll also generated a couple of times, I don't wait a long time. There is actually a parameter that you can add to your prompt that allows you to generate a few batches. You don't have to write it or copy and paste it yourself. Here in the parameter list, it's called repeat. You just can put this and that will create multiple jobs for a single prompt. Okay, let's see the results. Okay, this is interesting. I like the symbol here, but again we get the lotus. Let's see the other ones here. We have the candle, candle. Okay, maybe this one is pretty good. I like the candle holder. I don't know if that's it, but I assume it is. I want to generate more variations of this logo. I think it's simple and pretty. Okay. I still like this one way more. I'm just going to upscale it. Let's see if we can add more of the black aesthetic. Again, I'm going to try to blend this image with this one and see what's going to happen. You'd never know what the blend function as the images develop. I don't like any of them here. I still really like this one. But I'm not a fan of this flower here. What I'm going to do, I'm going to go to clip drop, then I'm going to use clean up tool and I'm going to upload my image here. Now I want to remove this flower. Okay, let's clean it. Removed it, but now it doesn't matter. I'll just save it. I'll go back to mid journey. I actually upload the images here. I'm going to upload the image. I'm going to use its image address. Imagine and then copy image address. And then I'll use the same prompt here. I'm thinking if I should use the other image as well. Okay, let's try to use that as well. So complete image address here. Let's paste it here. Now we have two images, okay, Spa logo featuring a candle surrounded by petals. Let's put minimalist and line design. Okay, style image weight. Let's make it higher, 1.2 I think that should be good. I'll just repeat it a few times. As I already said, you can use the repeat button here and put how many batches do you want? Let's put three more here. We are asked if we want to imagine three prompts. Yes, from all these generations. I like this one. I'm going to create more variations of the number for the first one here and the fourth one. I'm going to also generate some variations and then choose at the final logo. Okay, let's see, these are nice, like the first one here. Let's see, the other one here, I like number three. Okay, Here, I don't like any of them here. I want to choose the third one. I'm going to upscale it. Upscale number three here. Okay, I think it matches our description here. It's modern. It's capturing the tranquility and rejuvenation mood. And also we don't have too much details. The one thing I want to change is I want to get rid of this spiral thing. Here again, I'm going to go to clip drop. But first I need to save the image. Let's go back. Okay, I think that's good. Okay, beautiful. That's all I need. Let's download now. I'm going to go to Vectorize it and make a vector out of this image. Now let's use Vectorizer to convert our image to vector. Okay, great. Now we've got the vector, The quality is grade. The only problem with the background here, but it's easily removed. Let's download the SVG file grade. Now I can actually use Adobe Firefly to change the colors of my vector image. If I don't like the colors, I can try different color palettes with the generative recolor. The only problem here is that firefly is not available for commercial use yet. So I'm not sure how it plays out with the genes of free color because I'm here uploading my own image. I would be careful with that and just use that for inspiration. Let's say here, let's put black. Okay, now change to black and white. I like this here. Let's see other styles. Let's see the sandy stone. Let's use the lavender storm, the pink one here, Salmon sushi. Interesting here. Let me play around a little bit and choose the best one. I really like this one. I'm going to download it now because it has all this background. I'm going to go to online SG editor like this one. And I'm just going to upload my image now I will remove all the background here. Here is a final result after removing the background, which was very easy here. It's all really good and the lines are sharp. The only problem is this fire flame here, I, It just needs a little bit of tweaking and that's all here. The font that was generated by Mid Journey, I pretty like this font, so it can be used as the inspiration for actual company name. Here, this was our final project and you've made it to the end of the course. Congratulations, I hope you enjoyed the course. Found it useful. Please consider leaving a review for the course because it will help me to create more tailored content for you in the future. With that being said, I wish you best to flag with your AA art journey.