Stable Diffusion Deforum: Going Viral on TikTok & Instagram

1. Introduction: Hi, my name is RNA and in this course, I will teach you how to make this cool videos in the forum diffusion. So do you will not learn how to go viral on TikTok and Instagram. Although you just want to become an expert in stable diffusion and show your friends code videos, then this course is just right for you. In this course, we will learn about stable diffusion and how to use the forum to make videos that can go viral on social media. We will cover all important aspects of the parameters including camera access, camera rotations, sue, Right prompt engineering for stable diffusion and much, much more. You will learn how to properly structure the inputs to get the perfect videos to understand the full potential of stable diffusion, we also learned the necessary technical background knowledge. You will understand how a diffusion model works, what roles frames and seeds play, what key-frames during generators are, what they are, coherencies, and much, much more step-by-step instructions, you will be able to apply that learned knowledge perfectly in Brexit. Whether you're a beginner or unexperienced video producer. This course will give you the necessary tools to make good animations for the desired effects on social media. If you are ready to go viral or just want to learn, then see you in the class. There's no mercy again, this just hungry or thirsty. Parsons in different versions each do one question worsens why. We'll practice carton and you'll see the different Berman 2. What is Stable Diffusion: What is stable diffusion? The first question that comes up is maybe what's they will diffusion is stable. Diffusion is simply a diffusion model that's open-source. But what does it mean? Open-source means simply that everybody has access to it. Everybody can see the code and everybody can build on the code. The forum is one extension that somebody has made on stable diffusion. And the diffusion model is simply just something that makes your pictures, you gave it a prompt. Prompt is simply that the UPF, the diffusion model and the diffusion model gives you pictures. And we need to go to some of this background information because it will be important later in the course. Here behind me is a little Wikipedia post. And it says some basic things about stable diffusion. For us, it's only important that stable diffusion is diffusion model. You can find it on GitHub, it's written in Python and the rest you will learn in this course. So stable diffusion is diffusion model, just like meat journey and Dolly. But the big difference is that stable diffusion is open-source. Everybody has access to the code. Everybody can build on stable diffusion. And stable diffusion doesn't run in a Cloud. You need your own GPU to use stable diffusion, or you can borrow some GPU from Google or from runway. We will see all these things in detail in this course because I think it's smart to work with Google Colab. If you don't want to work with Google Colab, you need a lot of GPU on your computer, but it's possible. So what do you take from this video? From this video you'll take that stable diffusion, make pictures out of text, out of our brand. And that's it. 3. The Tecnologie behind Stable Diffusion: The Diffusion Model: As you know, stable diffusion is diffusion model. Who would have thought that? How does a diffusion model works? It's really simple if we make it simple. So just take a look at this picture. You see here a cat and here is the noise. And think of it just this way. You give the diffusion model a lot of pictures, possibly all pictures that you can find on the internet. You will give the diffusion model, cats, dogs, and everything that you know about and you will describe it. In example of this picture, we will say maybe a gray cat with a pink nose. And you describe everything that you see in the background where the focus is. Maybe also some camera details with what camera you have shot this and so on. And on this picture, we will add a little bit of noise, just like here, but with always the same text. And you go on with that until you have only noise left. In this process, the machine will learn how a cat, we'll look how this text will add up to something like that. And after that, if you give the machine only in the noise that text, the machine knows what this text means. This text means that you will see a cat. And if you do this long enough and often enough, the machine will basically understand how things look, what text will lead to what pictures, what prompt lead to what pictures. And that's important because we need to describe in the Bronx exactly what we want to see. We say in the prompt, only our cat, we will get a random cat because the machine doesn't know what Kat we want to see. We need to be specific. We want to have a gray cat that sits on the beach. For example, the sun is rising, the ocean is blue. And we can add some prompt magic eight k high detail and so on. We will look at the prompting later. But just a moment back to the diffusion model. I describe it like this. Just think of it this way. You'll look and this guy, and you'll see some clouds. Maybe your girlfriend stands right behind you. And Jesus, can you see the apple in the clouds? And you don't see an apple. But then she points at the clouds and say, Just look at it. It looks like Can I pull? And then you start to realize, you'll see on that ball in this guy, obviously there is no apple, but your girlfriend gave you the prompt. You are now primed to see an apple. Your brain knows how Apple's look like, and your brain now see the apple in the sky, even if there is no apple there. And that's exactly how a diffusion model works. Just broke down in simple, simple terms. But that's enough for us. 4. Why we need Stable Diffuison: Why do we need stable diffusion? I already told you that their style eat meat journey. And there's also Leonardo ai. And you can make really, really good images on these diffusion models. Leonardo ai is one diffusion model that runs on stable diffusion. So there's one example you need it to run other things on it. And the forum is one thing that runs on stable diffusion. And with different room you make this cool videos. So if you just want to make normal pictures, you will most likely use my journey dolly or Leonardo. Leonardo AI works also unstable diffusion and it's really, really easy to use. We will take a brief look at Leonardo also in this course. But we will use Leonardo most likely chest to steal some prompts. We can also steal prompt. We can use prompts, we can tweak prompts, and we will use them in the forum to make our videos. We will use these prompts, mostly from Leonardo and lexica. If you just want to make pictures used alley, meet, journey or Leonardo. But we need stable diffusion in the background. Stable diffusion is the backbone of a lot of cool features. Leonardo runs on it, the forum runs on it, and there are tons of other things that run on stable diffusion. You can even use other models and your own models to improve the quality of your pictures. We will see all of that in detail later in the course. So basically, we need stable diffusion to run things on it. And the forum is one thing that runs on stable diffusion and devoured them is awesome 5. What are Models, Leonardo AI as example: In the last video, you heard that unstable diffusion. You can run models. But what are models for that? We go in Leonardo and we'll take a brief look. I'm here in Leonardo and I'm on the fine-tuned models. But what does it mean? It means simply that stable diffusion is trained on specific pictures. So stable diffusion, in this case, the version 1.5 of stable diffusion, was trained on a specific data set of pictures to make pictures that look something like this. So if you have something in mind that looks similar to pictures of this, you should use a model that's specifically trained for this. You can also see that there's a training resolution. So this model is specifically trained to make pictures that look like this in our resolution of 640 times 832 pixels. And you can go on and on and look at different models. The Leonato diffusion model, for example, is drained, stable diffusion 2.1. It's made for resolutions of 1,024 by 1024 pixels, and it is known for incredible shaving and tie contrast. You make pictures that look like this. If you have something in mind that look like pixel art, you should use a model that looks like pixel art. If you want to make bad Lexus, maybe it's the best idea to take models that are specifically trained for bad Lexus. I think you get the idea. If you have something specific in mind, you can take a specific model for that. All this dancing style, this demons, and so on, that you'll see trending on TikTok. It would be maybe smart to use a model that is specifically trained to make this undead persons and the leonardo you have also another cool, cool feature. You can also make your own models. For example, here I have made two models. One model that I have trained specifically to make colorful flowers, and I have trained another model to make close-ups of insects. And I think in the future, you can also link your own models to default room, I think. And I hope Leonardo AI will give his API. And then you can link your own models, your own terrain models to D40. That will be a game changer If you could even train your girlfriend's face on a model and then link it to the forum and make videos. This specific person. So if you have a model that's trained on specific phases, you will get consistent output that will be important later, we will take a look how we can use models, and we will also look where we find models. We need to find models that we can download and upload in Google Drive. After the upload, we can use the models in default room. Most of them all these are for free on Hugging Face, and that's what we will not learn in this course. Later, we will download models, we will upload them in Google and we will use them in the ballroom. 6. What is the Seed, Midjourney as example: What is the seat and why we need it? The seed is simply something like this starting noise. That first signal that you're on diffusion model gets. If you give the diffusion model as specific seat, you can simply work more precise. You can tweak prompts and with the same seed, you will always get similar results. Just take a look at MIT journey and Leonardo first. This right here is a nice picture. We can see that prompts here, that negative prompts that will also be important later. And that right here, That's the seat. If we, for example, copy the seat and use the exec, same prompts, we will get the exact same picture. But if we use the same seed and simply tweak the brand a little bit, we will get a little bit different results. But thanks to the same starting noise, thanks to this seat, we will get really precise pictures. The picture will look just like the other or similar to the other. They can look at Germany and I will show you something. This dog's these two pictures right here. The right picture is simply simply a picture that I have made and the picture on the left is basically the same picture with the same seat, but I changed the prompts. So you can see you get pretty, pretty similar pictures, but they are a little bit different because we use the same seed, the same starting noise. We can tweak little, little things. This picture really similar to the other picture because of the same seat. But they can look at this picture. This picture right here, that's basically exactly the same picture as the previous one that I showed you. Because I use the same seed and the same prompt. So that same seat and the same prompt will give you exactly the same picture. And that's important. If we use us eat maybe a fixed date, we can work really, really precisely. We can tweak just little things. The same seed with the same prompt will give you basically the same picture. We want to have really, really consistent outputs. We need to work like that. And if we want to change chest little small things in the prompts, we will also get really consistent things. If we work with a fixed date, if we use random seeds, the outputs are more random. That's okay. That's totally fine. And I do that often. But if you want to fine-tune things, fixed it, make sense, or at least you should know the receipt. Later, we will learn more. But now you have a good basic understanding. What does the this and what it does and why we need it? 7. What are Frames: What our frames, for some, it's maybe common sense, but I need to make sure that we are all on the same page. Frame is simply one specific picture of a video. If you click pause my video and you'll see simply that one little picture of me. That's one frame of my video. You'll see something on a video camera like 30 FPS or 60 FPS. It simply means that the camera makes 30 or 60 pictures per minute. And all these pictures make up one video. Just think back on the first cartoons. What did people do? They use something like this. They draw some lines on the pages and they did this. All these pictures made a video and every page here is one frame example right behind me. Let's take this picture. I have made this picture in my journey, but you can imagine that this big check in real life would be in motion. So if you have a video of this, all these workers would work and not just stand here really lazy. But if you have justice that the picture, it's one frame, more frames made up a video. 12 FBS is 12 frames per seconds made up in a video. So just common sense that our frames and we need them 8. Prompt Engineering in Stable Diffuison: Prompt engineering in stable diffusion. As you know that Brown is just a text that you give stable diffusion to make your picture. And I already told you that it's important what text you give stable diffusion. Let's make a quick example here in Leonardo, right now, we work at stable diffusion version 1.5. Here we can choose our models. We take stable diffusion 1.5 for this example. Now here you can type in your prompt. Let's make an ugly little prompt. That's not perfect. The prompt is woman and press Generate. Now you have a picture of a woman, but the picture is not optimal. Because the prompt is not optimal, you need to be specific. You need to describe exactly what you want to see. Now I show you a good prompt. I'm thinking like this would work. A beautiful woman, masterpiece, realistic. 29 years old, beautiful face, dark hair, wearing a tight white shirts, studio lighting, cinematic light, eight K resolution. Whom? Human hands. Elegant, dynamic, highly detailed, and so on. Just be specific, GIF, Exactly, What do you want to see. Let's create this one. Now we have a picture that is more specific, but you may be see that the picture is not perfect, the hair is not perfect, the eyes are not perfect a lot than Nike's, maybe a little bit long. So we add a negative prompt. The negative prompt is the thing that you don't want to see in the picture. And in this example, we don't want to see a close up blurry eyes to face this plastic deformed, blurry but anatomic bad eyes, crossed eyes and so on. Just gave examples. What do you not want to see? Things like plastic, ugly, blurry, two phases, two hands, ugly, limps, text, and so on. That are all things that work is good because diffusion models, they usually don't make your texts. So in the negative prompt, you give the things that you don't want to see in the forum. It's not that important to use negative prompting because the videos are fast. So you don't see exactly what's going on. But It's important that you know, it's possible to use negative prompts. And if you want to make normal pictures, for example, in stable diffusion, you need to take negative prompts. Let's make this picture and you will see how much better the quality will be generate with the negative prompt. And now you see we have a real masterpiece. So negative prompts work really, really well in stable diffusion. First picture is a woman, and I must admit the woman is prettier than I would have thought normally, if you take a prompt that only says, Woman, you don't get really good results. After that, we go to prompt. But the last thing that missed in this prompt, whilst the negative prompt and I got prompt with a good negative prompt, you get really, really good results. Just look at this. Now we have a picture that looks really, really nice because of stable diffusion derived prompt and the right negative prompt. So in short, just describe exactly what you want to see. Use compositions. Use the resolutions. Maybe use cameras, use angles, use compositions. Describe exactly what you want to see. And in the negative prompt, describe what you don't want to see. Things like ugly blurry, two hats, two faces, deformed, ugly limps and so on. That are the classic things. Also long-neck, long body, just type in what you don't want to see. So basically, they'll stable diffusion while the woman z and what you don't want to see. We will need this later in the lesson. But I must admit the negative prompts are not that important thing in the forum, but you need to know them and we can copy a lot of prompts. And in the next video, I will show you where you can get some inspiration for your own prompting. 9. Finding Inspiration on Lexica and Leonaro AI: Getting inspiration and maybe also some prompts. In this lecture, I will show you where you can get inspiration for your own prompts or where you can copy them. You should go to showcases. One showcase is Leonardo, and I liked this showcase really, really much. You can take simply pictures that you like, maybe something like this. And you'll see that prompt and the negative prompt, this things you can copy. Another showcase is lexica and tear is basically the same thing. And you can search while you will not see. Maybe you want to see yeah, demon, just search it and search for the picture that you like. Click on it and press Copy prompt. If you want to copy them. You can also take likes right here. And after that, you can go to your likes. And you will see some picture that you already like. And you can simply copy the bronze of these pictures. Maybe you like this venom and you can simply copy the prompt if it's right for you. Same thing in Leonardo. Same thing in leonardo, you can simply search for something, maybe a zombie, and you will get mixtures, press on them and copy the prompts if you'll think they are nice. And now I want to show you just a quick, quick what the z thus press on this picture. We have already seen that, but I want to show it again because that's important. You can hear re-mix exactly this picture by pressing on this button. Now we generate this. As you see, we have two pictures that are pretty similar to this one. But what's the difference? The pictures are similar but they are not the same. But what if we simply copy also the seat? Just copy the seat. Will let the bronze and everything else the same. Now we use a fixed date. We activate this eat, we booked the seat right here and we press generate again. Now you'll see the picture is even more similar. This picture right here, you can say that is pretty much the same picture or stays right here. So the seed is really powerful in my mind, you can see the colors and everything they match, whether to this picture than the previous one. Now you know where you can get inspiration from. Use Leonardo, use lexica, both of these prompts, negative prompts. You can use both of them. And you also seen the seat in action 10. Join the Deforum Discord Community: Join the deformed community on this card if you want to, always be up to date, Of course, you can also join the community and you will see immediately what's going on in the forum. Simply press this right here. After that, you will have this default room in your discord. And of course, here is everything you need to know. They make announcements. Sometimes if there are cool things that are happening, They have all the official links right here for you. They have a quick user guide. They have some tutorials, and you can simply chat with other users. My mind, it's really worth to join that community on this cart. Of course, you will always be up to date in this course, but a community is always nice. 11. Local Installation of Stable Difusion with Python: How you can get the forum or and stable diffusion. One method is to install it locally, but that's not the method that I liked the most. But if you have a really, really strong graphic card with 16 GPU or more, you can do that. In my mind, it's maybe best if you have a little, little bit experiencing biotin. I just go to a simple and quick, quick overview how to do it. But I must admit that's not the thing that I do. After that, I show you two videos, how you can use the ballroom without a lot of GPU. And in my mind, it's the better way. So you need to download for Windows. You need an account in Hugging Face, and you'll need to download that automatic 11. 11. You need Biden installed on your computer. And you need the Hugging Face account and attacking phase. You need to download the stable diffusion models. You can start with the model 1.5 anthem all the point though. And after that, you can also include other malls if you like. So if you have a really, really strong graphic card and maybe some experience in bitumen, you can download all of this and granite locally on your computer. The next two videos are for you. If you are just like me and you're a little bit lazy, don't want to install all the things and run all on the Cloud. I also have to say if you install this locally, all your GPU will get used to render your videos. So as soon as you start rendering, you will most likely not do a lot with your computer, expect their rendering. The next solutions are in my mind that better solutions. And with the next dilutions we will work in this course. 12. Rundiffuison: A Cloud solution that works: The second option that you can consider is random fusion. Random fusion borrows you there GPUs and you can use them to render the videos. They have all installed. And you can simply click and run it or you can simply go to app dot run diffusion.com and just dry it. They have different GPUs for you. They have HE abide ramps, they have 16, and they have 24. Use the subscription that is perfect for you. They borrow you there GPUs for $0.50 an hour up to two bucks and $0.50 an hour. And it's right. You need to pay that per hour, and it doesn't matter if you render or not. So if you work in stable diffusion, you will need to pay at least $0.50 per hour. That can add up, that can be expensive, but everything is installed also automatically 11, 11. So you can also work in default room. You can simply borrow their GPU. Everything is installed, everything is ready, but you need to pay $0.50 an hour. And in the next video, I will show you Google Colab. In my mind, That's the best experience. You can get the GPUs from Google Colab. They are really, really cheap. You can connect and disconnect really fast. So when we try to prompt and make our decisions, how we make the video, we can simply disconnect, we connect to render, and it's really, really cheap. In my mind, Google Colab is the way to go and that what we will learn in this course. But if you want to try around diffusion, feel free to try it 13. Colab Notebook and Google Drive: The Google Colab Notebooks, in my mind, that's the way to go. Google Colab is really nice for default room and everything you need is this copy and a bit of space in your Google Drive for the models, I think everybody has a Google account. So simply copy this notebook in your Google Drive and do everything that's necessary. I must admit that it's possible that you need to make a subscription. Subscription costs $10 a month. You can try it if you like it, go if you really hate it and don't want to work with the forum, just cancel it. And if you have some luck, you can try it for free. It depends a little bit how many GPUs Google Colab has free right now. This right here is the interface, if you wanna call it interface. But we will take a close look at everything that's necessary to make the right decisions for the forum. Even if it's a little bit overwhelming right now, you will understand really, really quick how all of this work. If you wonder why my page is black, they'll simply two tools, settings, and make the theme either adaptive, if you hold Windows, is dark or take here the dark mode. I have, my whole window's a dark mode, so that works fine for me. Right now We work in the forums, stable diffusion version 0.7, every now and then, new versions pop up. But it's simple. Just copy and new page of Google Colab notebook and you will have the newest version. The next thing that you have to do is simply to copy this in your drive. And don't worry that these words are in German. I'm from ethylene speak German, but in your case, in standoff in Google Drive, copy it and there will simply be the word copy in Drive. So press to copy in Drive. If you go to your Google Drive, you will most likely have this order now, Colab Notebooks. If you make your double-click on it, you will most likely also have just one copy of this thing right in here. And you can make copies of this copy just to make sure that we have almost always a fresh notebook here in the default settings. If you double-click on one of these, you will simply land in your copy. You can rename the copy. And I will do that right now and make this example. And we can simply save this as example. After that, you need to click Run and Run all. Now everything will get installed. Then models will download automatically in your Google Drive. And after that, everything is ready to go. You need to go through this process just one time. You need to download all models. All everything needs to be installed and everything needs to be in place. After a few minutes, you need to give permission to your Google Drive. Just take your Google account. Gave permission. This whole notebook is designed that way that everything will run automatically. So first this cell, then this cell, and every cell after that, we'll run automatically. And I think that will maybe take you some five or 10 min until every cell is done. If we go down a bit, here are, for example, the prompt cells that are the prompts that you will later gift to the forum to make your animations. And right now we have chose to animation mode, none. That is the default setting. And if we take none, this prompt, so they sold their normal prompts will gather around and these two pictures will get created automatically as soon as this notebook has done its thing. You also see that animation prompt right here. And with the animation prompts, we will make our videos. And for the animations we have to choose something other. If we take none, we will get pictures. If we take to the 3D video input or this right here, we will get other outputs. The animation prompts and the videos are for other electrons. If you scroll down right here, now we will get our first pictures because we have chosen none. We will get two pictures here of the prompts that are in the default settings. And you'll see it right there. These are our first two pictures. Because we chose none. We get these two pictures. So a beautiful lake and the portrait of the woman. Everything is set, everything is ready to go. These two pictures will now be in your Google Drive. If you go to your drive, there will be a folder AI. If you double-click on it, you will have here most likely that folder stable diffusion. If you click on it, you will have this folder and there will be a folder stable thumb. In the folder, stable fund, you will have your pictures right here. And right now there are a lot of other pictures that I already made with stable diffusion. But in your case, if it's your first time, you will only have these two pictures that we created right now. Everything is that everything is ready to go. And in the next video, we will take a look at the pricing. If you want to make more animations and guilt animations, most likely you will have to make a little subscription 14. The Pricing in Google Colab: Let's take a quick look at the pricing. If you are a new Google Colab notebook, you will see that they gave you a video card. So the Tesla T4 is right now the video card that we got. And if you're lucky, you get this for free. But if you're not lucky or there are a lot of people that are also want to have good video card you need to make as subscription. You'll simply press on this right here. Right now you can see that I have 60 h left and I will pay 2 h bare. Our basically just press on this right here, and you will see the following. You can make as subscription for Google Colab Pro, and it will cost, in your case, most likely $10 a month. You can also buy the Colab ProPlus body. In my mind, you really don't need this. They collapse zero gives you 100 h. And in 100 h you can do a lot of stuff. You have 100 h to render your videos. And with this graphic card, we will pay 2 h/h. So we have 50 h to render our videos. And that's really a lot. 50 h. You can render a lot of videos and you can always disconnect. If you don't wanna render. If you try to prompt and you'll make your settings, you can disconnect and it will don't cost you your money. Don't cost you a European beautiful units. As soon as you are connected, you will pay 2 h hour and you can press simply on this right here. If you want to disconnect, click simply the dumpster right here and cancel it. And right now we don't pay nothing. We have our 60 h left and we will pay nothing. We can make our adjustments right here, and we can simply connect as soon as we will now create something. So we might world $10 a month are unacceptable price because you get your GPUs, your computer will not run hot. You can do other stuff with your computer and 50 h of pure rendering Israeli a lot. It's much cheaper than the run diffusion. It's much cheaper than other solutions 15. The Path Setup for the Models and Output: The best setup. In this video, we will take a brief look where your models are and where your pictures and your animations will get saved. Path set up. For me, there are only three things important, the model path and the output path. And you can see the model path goes like this. Content drive My Drive, ai models and the output path goals content drive my drive a stable family. If you go into your arrive and simply see all the things right here. So drives my drive. Now we go into AI, drive My Drive AI. Now stable diffusion or models, these folders I have created on my own, we will do that later. You have most likely these two folders models right here, there are your mother's, most likely you will have just one model right here. And under stable diffusion, this right here, stable fun, there are your outputs. So really, really simple. Models are right here and the outputs are there. We have also the model setup and we can take our own models right here. We will learn how to upload our own models and how to use them later. Model checkpoint, you'll need a model that you have in your drive. For us, it's most likely this right here, and on the default settings, it will be right this later we will maybe choose a custom model to make some more pictures. In my mind, That's it with the models. We hold this right here. Really, really simple 16. How Deforum works: What is the forum diffusion? With the forum, you simply create frames. You know what frames are. We can create frames. We can create 50 frames, 100s or 1000s frames. We make all frames that we want to see and we can adjust, ever reframe how we want to see. We can include seeds and other stuff. And we can also include camera access, rotations and everything that you can possibly think of. But the only thing is that we need to adjust all these things accordingly to what we want to see. We need to be specific. In the following lectures, we will take a close look how that works. So the forum is basically just stable diffusion. So diffusion model that creates pictures or frames of a video. As soon as we have all our frames that we want to have, we can blend them into a video and we can use pictures of our cell, for example, to start with our first frame. After that, we can transform this picture in everything that we want to see with this diffusion model. So to summarize this, the volume diffusion is an extension that runs on stable diffusion and it can create forams for you. You can make these frames how you want to see them. You can input your own videos. You can input frame or a picture of yourself to start with, and you can deform your frames, your pictures. And everybody works in the forum like this. You give the parameters that you think are right. You click Render. You see if it's cool. If the output is good, alright? If the output is not good, you tweak parameters a little bit and your render again, that is how you work in the ballroom. And in the coming lectures, we will take a closer look at everything. Let's start with a brief overview in the next video. 17. Overview and Quick Guide: That the forum quick guide and overview. In this lecture, I will give you a quick, quick overview that you have something in your mind. Because after that lecture we will look at everything in detail. I have found this quick guide really, really helpful. And as soon as you are done with this course, you can make animations just like this with rotations in 3D and so on. But first we have to take a look at the basics. You can make your animations in to the end in 3D. And it's important to understand that if you take none, you can also make just the pictures. But we look now at the 2D and 3D settings where you have the emotion side things in 2D. The first thing, it's the Zoom that right here is assume of 0.99 and it zooms out. And the same thing, assume of 1.01. It's in place, zooms in, as you see in this animation. That 2D angle rotates your camera. 2d translation x will move your camera to the left or to the right. Positive values, we'll bring it to the right and negative values will bring it to the left. That translation y, we'll move your camera up and down in the same manner. Positive, bring it up and negative values bring the camera down. If you like. With this value, you can also zoom in specific corners. But that's something that I usually don't use that much. Besides that, you have also the to the perspective flip and that looks like this. Now we come to the 3D motion settings. Translation x is basically the same thing as in to the ends. Always translation, y, translation, that is in 3D, the same thing as the zoom into the bacteria work with higher values. There, free rotation, rotates the camera up or down. 3d rotation y to the right or to the left. And the 3D rotation, rotate the camera positive values clockwise, negative values counterclockwise. That was a quick twitch twitch overview. And it would make sense to open this quick guide in a window and use it as soon as you start typing in your own commands. Because it can be a little bit tricky if you start with this. So just open this quick guide and leave it in a corner where you can see it. And now we dive into more detail 18. The Basics of Animations: Angle and Zoom as example: Animations and set things in the forum. In this video, you will get the basics, that stuff right here. We have all done. Now with the sad things. It works like this. Animations. You know, with none we make our pictures and 2D, 3D or video inputs are four videos. For now we worked with 2D to make it a little bit easier. The next thing are the max frames. How many frames do you want for your animation? Let's say we want to have 200 frames for our animation right now. Border replicate, we can simply leave this editors. The first thing right here on our emotion parameters is the angle. And you know, from the quick guide, the angular rotates your animation. The first thing right here, that's the frame. So we start at frame zero. And the thing in the bracket is the value. So we don't rotate at all. If we put one right here, we rotate the whole video. So from frame zero in the value of one. And we can animate this. If we make a comma, we can do something like this. For example, from frame 52 points, we want to see rotation that goes counterclockwise, so a minus of two. And we can animate this further from frame 100, we want to have a rotation that is again clockwise, but, but this time free, that's a fast rotation. From frame 150. We want to have, again some things that I think, you know, what do you get from frame zero to 50, we have a rotation that goes clockwise until frame 50. At frame 50, we start rotate counterclockwise at faster speed. At frame 100, we start rotate clockwise again and we rotate really fast. And DataFrame 150 after 150, we are static with the zero. That works in all of this. You can animate every frame, for example there. So we start here at frame zero again, we have assume that zooms in, as you know from the quick guide, we make a coma. And let's say from frame 50 ohm, we wanna have assumed that does absolutely nothing. So this right here is static. From frame 100. We want to have assumed that does something other. For example, let's say we want to zoom in but really slow. And don't forget to put your commerce. And at frame 150, we do, let's say negative 0.99. And that's how we animated this thing right here. We did the angle first rotation clockwise, then counterclockwise, clockwise, but faster. After that, the last 50 frames and DataFrame 200 because we build tier 200 statics. The Zoom. First we have a pretty fast to my DataFrame, 50, 50-100. We are static. 100 to 150. We have a Zoom again, but a bit slower. And 150 to 200 we have a negative x2. That are the basics. You can animate every single frame of this thing on every single parameter that was angled and Zoom. And in the coming lectures we will take a look at everything. And the basics always stay the same. You work with frames, two points, and then in the bracket is the value. How big the value is? You should know from the quick guide 19. Kamera movement in 2D and 3D: Translation, x translation y translation c. And of course, their rotations. These things really need translation. So, you know it also from the quick guide, the x-axis is simple. What does the x-axis? The x-axis moves simply the camera. It moves the camera to the right or to the left. Positive values will move the camera to the right. Negative ones will move the camera to the left. And we can work here with values from negative eight to positive eight. In my mind, that's more than enough. And same thing is for the y-axis. The y-axis moves the camera up if you use positive values. Same thing here where you can go up to eight and negative values will move the camera down. I recommend you to play with values that are two or three to start with. You can go up until then, but in my mind, 10,000 bring the best results. So basically x2, this, y, lose that. And you already saw their rotations and the rotation make the most fun. I find the best way to explain it is this. So the first thing is take your right hand and try to do something like this on the picture. Translation x will rotate your camera like this. Translation y like that. And translation set will rotate your camera through this. I hope this all makes sense. And I think pictures do that better chop. So now I will blend you again the pictures from the quick guide in my mind, if you see something that that's always the best thing, just look at the pictures, how they rotate, and you will understand how the camera settings will work. Positive values again versus negative values. You can use both of them. And remember they're free the rotation right? Now. We don't use them. We work with 2D, but later we will use them that are basically the rotations and the camera movements. And don't be shy. If you do something wrong, it's no problem. You'll make some adjustments and that, and maybe it's even better the way you didn't want it first. So I think now you have a basic understanding of the translation x, translation y anti-free the rotations and as you know, translation, that is the Zoom. But in 3D, under rotations, you work with smaller values. The values in the bracket should not be higher than three. We will look at that later. So translation x, translation y, we worked with a little bit higher values. That, that is the zoom and the rotations simply rotate. I think now you have a basic understanding what these camera access to. But here is a formula in the bracket. And at this formula, we will look at the next video. 20. What does the formula in the brackat mean: Formula in the bracket, what does it do? You can also use formulas to animate your friends. You don't have to type every frame yourself. You can also use formulas and you can even use KEY string generators. And also at the generators, we will take a closer look. But first, what's this formula? This formula simply animate your camera movement. But how exactly will this formula animate it? We can ask the chat GPD, what will this formula do on our x-axis? It will simply make waves. So it amplitude a wave with the value of ten. So in this case, the wave will reach a maximum of plus ten and that minimum of minus ten. And you know, that's what we do in the forum. But how we can make it even simpler, we can go into Google and simply copy our formula. And now you see it. We move on the x-axis constantly, from minus ten to plus ten, minus ten to plus ten minus ten plus ten. So on the x-axis, we simply go in waves all the time. And with such formulas, you can make a little bit of time segment. And you can put everything that you want, the nose you can make through casinos for example. You can also make this Tanga and if you want, and you can animate this further. You can also say maybe casinos and then dimes, dimes, something other times this right here, but this time, see you knows, you can play with these brackets how you want. You can put it in Google and look how the camera behaves, but you don't have to do this. I normally don't use that much. But if you like, you can use formulas to make faster animations. If you want to have a specific camera movement, maybe a wave that goes all the way. You can try this. And of course you can also adjust the values. So instead of free, maybe make here an eight and you will see the camera will behave completely different. That derived here, we'll give a really, really cool animation in my mind, you can simply play with these formulas how you want. And if you find something cool, just copy and paste it right here in your bracket. And you can even make a coma and animate from frame 100 completely different. Let's say first you will not have this animation. And from frame 100, you want to have the cool animation that we did right here with the camera movement that behaves a little bit silly. That's it. So first, we have the camera movement from frame zero to 100 in normal waves. Later on, we have some really funny, funny camera movement, or you can simply delete all of these and booked zero. If you put zero, we will be static. But make sure to put this also in brackets. Of course, after our frame, we need to put this formula in a bracket. And this right here, we don't need that space and that's it. So I think now you can understand how these translations and rotations work. You can put everything on every camera movement. You can animate the frames by hand, by KEY string generators or by formulas. And for the key string generators that for later lessons. And you can always put zero if you want to be static or a simple letter in the bracket. For example, on translation, why we want to move into speed of three positive. And that's it. 21. Noise schedule and strength schedule: Noise and strength, that are two things that I will not explain your next. Now you can see their perspective and the perspective flips. You have seen them in the quick guide. I personally don't use them, but this right here, this is important. I also don't use them that much because it normally works only in 2D. But the noise casual is really important and so is the strength. What is the noise? As you know, stable diffusion is diffusion model. It creates the things out of the noise. And if you take a lot of noise, it can make the big chairs faster. If you want to make an animation where this Zoom is really fast, it would make sense to give stable diffusion a little bit more noise. Stable diffusion creates your frames, are your pictures out of the noise. And if the Zoom is too fast, stable diffusion doesn't keep up with generating new stuff. If you make it a little bit noisier, stable diffusion will have an easier time keeping up, creating new stuff. So are really fast too. You could use a higher noise. And the next thing is the strength. What is the strength? Here? You have a strength of 0.65. And remember, you can also animate all of this. So let's say you want to have the noise from frame 100. You want to have it a little bit higher. So 0.04 from frame 100, you can animate all of these just like we have seen. And same thing is with this strength or this strength says the following. You create a frame and the next frame will be 65 per cent, similar to the previous frame. If you put here 0.80, for example, the foramen will be 80 per cent similar to the previous frame. If you take 0.12, it will be 12 per cent similar to the previous frame. So you can play with these values a little bit. If you take it too high. If you take it one, then video will not animate because all frames will look similar. If you take 0.9, for example, you will have a pretty, pretty consistent outputs. And 65 is the default setting that works really well for most cases. But let's just make a quick example here with the noise. Let's take our Zoom. First, we have assumed that a little bit faster. And then after frame 50, we have a really slow or near static soon. So it would make sense to increase the noise a tiny bit. First. And then after frame 50, we could use the standard noise of 0.02. And the strength that works for me just fine. 0.65 works for all frames in our case. So just remember, if your resume is fast, give it a little bit more noise. You don t have to do that, but you can try. And the strength, you can increase or decrease it just like you like, just remember that first frame and the second frame and so on. They will look similar to the previous frame. They will look, in this example, 65 per cent similar to the previous frame. You can play with that. I recommend 0.4 to 0.8 because higher values and lower values spring, sometimes uglier results, but I recommend playing with this. And there is also a relationship between the noise and the strength. If you use higher strength values, you can use less noise. They can look at this example. There is a lot of math behind of this. Enter basic suggestions are that if you increase the strength to 0.9, you can or should use your own noise. If you go down to 0.7 strength, you can use a noise of 0.02 and the strength of 0.5, you could use a noise of 0.04 that are rough estimates that are starting points and you need to play with them. And we have the right noise for our strength. In this example, we have 0.65 strength and our noise is 0.0, 3.0, 0.02. And that is right in between of these two 22. Coherence and cadence make the videos smoother: You may ask yourself what's next, and I will tell you that coherence is next. This stuff right here. I like to leave this as it is because in my mind, these changes little to your outputs. You can also use a mask and the anti blur. And I personally also don't use this much. But the next thing right here that's important in my mind is the coherence. You can use the color coherence and you can simply try what's best in your style. I have found that HSV is in most cases relatively good. But you can also work with lab or the other ones. These differences are just minor. They make a little bit of other colors. And this thing right here, that is important, diffusion cadence, I use here nearly every time. One, if you use one, that if your model will diffuse or generate every frame, if you use two, it will only generate every second frame, and it will blend A-frame in-between the two frames. And if you take eight, of course, that will take a lot of render time. And it will blend eight frames in-between the two frames that will be diffused. Just try to play a bit with this. If you take higher numbers, you should get us move for video. But in my mind, it's best to work with one or two, but trust, simply try it and see what you get out of it. I, personally, I use one in this guide right here. They describe decoherence like this. So of course you can play with all of this LAB HSV and so on. And the example of the cadence five is this. One foramen will get diffused and there will be blended or drawn for non defused frames in-between of debt. So the total output is five. If you take eight, the total output is eight and only one frame will diffuse. Just simply try to play a little bit with this. I personally, I use just one. Normally 23. UPDATE: Version 0.7.1 is here!: The AI space moves really fast, and for that reason, I'm also fast with my updates. Here in the forum difusion, there was a small update. We are now in version 0.7 0.1 The community said that the prompt box was a little bit too complicated, and for that reason, the prompt box is now a bit simpler. Let's just take a look. Every other stuff right here is exactly the same. But the prompt box got a little upgrade. If we scroll down, you can see right now there is only prompt negative prompts. And the negative prompts will work. Just as always, you don't need animation prompts anymore. You can type in your simple prompts right here, even if you want to make pictures or video that makes no difference. You can also include your normal prompts right here. You can include the hashtag to exclude something. Everything else stays the same. So you just to take into account that and the negative prompts work without animation prompts or something. It all depends on what you use right here. If you take non, you make pictures and your videos you make in 2d3d or video input. If you scroll down a bit, here was patch settings and over patch settings there was also prompt settings. Also the prompt settings are gone to make it easier. You also don't need to make prompt settings. Everything gets easier, everything is more. Right now, simply try it out. 24. The prompting cell and negative prompts: The next thing I want to show you is the prompting cell. And in the prompting cell, you will enter your prompts. If you scroll down a bit, you will get to the prompting cell. And here are your normal prompts. Here you can simply type all the prompts from what you want to make a picture. Every time you make a new one, you can simply put this right here and create your prompt. For example, a dog or whatever you want to see. And now we will make free pictures of these three prompts. But just if we take this right here, we need to choose none. If you want to activate this bronze, if you want to activate one of these prompt, simply put this right here in front of it. And when they are agreeing, they will get totally ignored. But here I want to show you that animation prompts. That animation prompt will get activated if you choose something like 2D, 3D or video input. If you choose something from the other, these prompts will get activated. How this works. Now, from frame zero up to 20, you will see our beautiful upward trending on art station. From frame 20 to 30, you will see the banana, and from frame for the 240, you will see your coconut. And from 40 until the last frame, you will see this thing right here. Same thing right here. If you want to activate one of these, just put this before the prompt and it will get green and it will get totally ignored. So in this example, we have our banana until frame 41, last little example that Apple will be an apple until frame 20. And from frame 20 on the apple will deform in a banana, and that will happen in a few frames. Your strength up there will determine how fast the apple will change into a banana. Frame 20 in this example would be 65 per cent an apple. But the frame after that, it will be simply a banana. So I think you get the idea from frame zero to 20. You have an apple. From 20, you've heard that you have the banana. Banana, banana has to transform. And that's based on the strength that you gave them all. 65 per cent of the first frame will be an apple, but the frame after that you will have nearly only that banana. If you increase the strength, it will go slower. If you decrease it, it will be faster. And you can change and go on with what you like. You can also make the banana, for example, 200 frames long or how long you want. And you can of course, make just u1 and say, let's say from frame 60, you want to see a cat. You can do whatever you like with this prompts. And as you know, in stable diffusion, we can also work with negative prompts. And the negative prompts work like this. Let's say you have the cat. Let's add a little bit. Prompt magic eight K, high-quality, vibrant color. And after that, you do it like this. Minus, minus neck. That are now our negative prompts. And you'll just type in the things that you don't want to see. Ugly, blurry, extra limps, deformed. Answer one here or you can simply put what you don't want to see. And of course, you can simply copy the prompts from Leonardo. This right here is just an example. Just go in Leonardo, copy the positive prompt, type in minus, minus neck, and then copied and negative prompt. And you will have really good prompts for stable diffusional. One morpheme. You can also include the negative prompts like this. You can simply type of income. Dash, negative prompt equals bracket, make a new one, and just type something like this. Ugly. Also these you can put wherever you want it. You can also say from frame to frame 20, you don't want to see blurry. You can do whatever you like also with this negative prompts. But I must admit I nearly always use this negative prompts because with this, I have some issues sometimes, so I delete them right now. That whilst the prompting cell, if you choose none, you will get your pictures. If you choose to 3D or something others, you will get your animation prompt. You can type on every frame, what do you want to see? And you can include negative prompts. Just simply put minus, minus neck and type in your negative prompt. That is the fastest solution. And if you don't want to see something, booked a hashtag in front of it and it's gone 25. Load and Image settings and the resolution: Load and image settings at these two things, we will look in this video. Underneath the brown box, there are the load side things. These first things right here, I usually just leave them as they are, but this will be important. Image settings. What does this mean? Width and height. It's simply the resolution. So we get that picture, 512 by 512 pixels. Why we have here $500 by 512, because our model was trained on this resolution. So in theory, we get the best output if we use our resolution of five-twelfths by five-twelfths. But most of you will most likely make pictures or animations for Instagram. For Instagram, we have to take other resolutions. There is no way around it. For instant TikTok, there are other resolutions. Five-sevenths is 61,000.24 works optimal for n star and TikTok. So if you want wanna make things for your phone, you should use debt resolution. And I will also link to a PDF where you can control what resolution for what you should take. But for our example, right now, we want that simply roll with 512 by 512. Because for us, it makes no difference. For me right now, it's only practical to take lower resolutions because the render time will be faster. And this right here, Bit Depth output. You can leave that on eight or you can increase it and 32. But what does this mean? I meant by example. You can also double-click on this right here and you'll see the code. And if you are a coder or you can do a lot of stuff right here. I personally, I'm not a coder, so I double-click again and it goes away. But if you are a coder, I'm sure you can do a lot of stuff here. But back to the bit depth output. If you increase that, you simply get more colors, you get deeper colors, you'll get more colors. You will be more precise with your outputs, but the render time will go up. And I think you can or should start with eight, safe some render time. And I think the resolutions good enough with eight. So in simple terms, what we learned in this video, bit depth output, leave it at eight. If you want to increase it, you will be more precise and get more colors. And the image settings, you need to take the right image settings. For me, 512 by 512 works wonders because it goes faster to render in this course. But for you, maybe 576 by 1024 with work really good. If you want to make things for TikTok can then stop 26. Sampling Settings, the seed and the steps: That sample site things. And now we get to the seat. You can see it is right here under the sample settings. There is the seat. Right now we have a seat of minus one. What does this minus one mean? It means that every picture will get on you random seat and you can't track it. So every seat will be completely random. The first seed Cook be five, the next seat Cook be 53,000 or 2 billion because the seed is really, really big. There are a lot of different seeds out there. If you want to be more specific, you should put here our seat. What do you use is completely irrelevant. It could be one, it could be ten. It could be, for example, 888. It's just important that you can track your seat because eventually you will need it. Under the seat you have the sampler. There are different models. Some of them are a bit faster, some of them are a bit slower. I like that default settings are the first one right here. But for now we just work with the default settings. The next step right here is steps. But what does this number mean? It means that the AI, that stable diffusion will run 50 times over your picture. So it controls 50 times. If your picture is right or not. You can use here higher or lower numbers, but something 40-70 works pretty well. And I must admit I use mostly the default settings because the default settings work really, really well. After that, you'll have the scale. And just like in leonardo, you can use a higher or lower scale. If you use a higher scale, stable diffusion will lean heavier into your prompts. Something 5-15 is normal. I usually also use here the default settings. But in some animations, scale of ten makes sense. If you want to have really specific prompts as Kayla, sudden is maybe a bit too low. I personally, I work 7-10 most of the time. And of course, back to the steps just for a moment. If you increase the steps to 70, for example, your render time will go up. Safety will most likely be enough in most cases. And these things right here, until the prompt settings just leave them as they are. The default settings worked fine in most, most cases. So we'll leave them. We need to go on with our course. In the next lecture, we will look at the prompt settings. So in this lecture, we took a look at this seat. Use a random number, but you should use a number. Don't use minus one because our random seed is not optimal scale, you'll seven to start and use 50 steps to start. That default settings are fine here 27. Batch settings and prompt weights: Prompt and bedside things. And I know there's a lot to talk about, but that's the way it is in the ballroom. So right now we are in the prompt settings. You could use prompt weight things, and that's one thing that I will show you right now. You can use something like two points and two or three or something, and you will get more strength on that specific word. For example, a beautiful green apple. And if you put two points and two, the green will be twice as much importance as the apple. And if you put three, it will be three times more important than the apple. But I must admit I most likely don't use that at all. But you can, if you want to use it, if you make a prompt and you'll think one specific area of dead prompt should be more neglected or stronger, or you can simply put two points and R2 or R3 behind it. And right here in that prompt settings, you can activate it. So leave it as it is. Normalized. Prompt waits, just leave it and Locke waited. So prompts, I don't use them at all. Now, down to the batch set things, this number right here is simply how many variations you get. We believe that I do own right now, the batch side things. That's where your pictures will be saved later in your drive. So let's just make an example for this and simply call it banana file format name. I use mostly the seat. If you take really, really long prompt right here, your name will be crazy. Behavior. The default setting is either, what does either mean right now we have a set of 888. Our first frame will be rendered at 888. If you take their second frame, will be rendered at the seat of 889. After that, the third seat, 890, 891, 892, and so on. With the ITRC, you get a really pretty consistent outputs because they see it alternates only one at a time. If you use that fixed it exactly. You have the fixed date you have on every frame, the seat 888. And that makes your output even more stable, more similar. If you take a random seat, exactly, a random seed will be applied. If you want to make something that's really funny or really don't need any specific look. You can also use random. After that you have leather and that simply goes, I think two steps up and then one down and two up and down and so on. And an alternate that something in-between of leather and random, I think, but I must admit I use most of the time iter and fixed that are the seats that I use normally. I don't know why we even have a leather. I don't know it, man. So for this example we use either seat, either number. You can also increase the item number. So in our example, the item number is one. That means every frame increases the seed pi one. And we don't want to make grits. And right now that's it for the prompt settings. Just remember how your seat will behave and where we are. Images will be saved 28. The last settings befor the rendering: In this video, we will look at the last settings and in the next lecture we will go render our first video. So now to the left side, things in IT settings, what our image settings. If you want to take your picture, you to start your animation, just like in TikTok, we will need to use them, but we get later to this. For now, we'll leave it alone. Under the init settings there, There's also the mask file. And if you want to mask a specific thing out of a video, you can also do that. But right now we don't use any mask or any innate setting exposure. In contrast, I only used the default settings right here because I have found that there is little to no difference if we make here something. And same thing with match conditional settings. We leave this all at the default settings. And I have to say all the rest, we simply leave it at the default settings. If we scroll down, we have here our two pictures, our output pictures, our video will appear here, but only if we disconnect this right here is keep video for ran all. If we click on it, we will skip the video. If we unclick this, we will make our video. But you can also run all the other things. Leave this as it is, and make the video afterwards. And if you work with Blender, for example, or After Effects, you can also use all your frames and simply put them in Blender or After Effects. Most of the time, I do that in Blender. But for this course, we will take this step right here. So we uncheck this box and T or you can make your frames per second. So if you want to have the video with two other frames per second, just leave it at the default settings. 24 frames per seconds are also something that we see often and you can also use 30 s. Just use a normal random number. If you have a camera, you can also use the same frames per second to make your videos if you want to blend them together. And here you have only the paths left. So wherever your stuff will be saved, our video will be saved under Dr. MyDrive, a ICE table diffusion and stable fan. And the same thing for our video later on. The last thing right here is the max frames. How many frames max do you wanna render your video? We have here through handwriting, the default settings, and I think our animation has also non-modern through hundreds. So yes, that works. If you put here 250 and you make 250 frames, you will also need to increase your video right here, or the last 50 frames will not be rendered. You can increase this how much you want. That was it that are the last settings. In the next lecture, we will render our first video and we will see if we have luck, if we can use the free credits from Google Colab or if we have to subscribe, I already have a subscription, but you can try it for free. I wish you the best luck. 29. Render your first video: It's render time. But first, let's take a brief look, wild with it. This right here. If you work normally in stable diffusion, it's all set and done. But we need to play all cells after that because in the course we are a little bit slow with our adjustments. That animation we take to the max frames, we have a 200, the angle, the zoom into translation x. We did that all together, so we rotate clockwise and counterclockwise, clockwise then static answer one. The next thing is the prompt. Prompts go until frame AT. And we have first our apple banana than our cocoanut. But I will not increase the frame number here a bit. So maybe we take these 230 right here to 60, these to say 100.150. We wanna have our cat resolution, 512 by 512. And all the other stuff is okay. And I disconnected this right here, so I uncheck that box and we make a video with 12 frames of it. And we need all in all 300 frames. And that's it. You click and run and run all. Because most likely you are also a little bit slow on your first time. So we need to run all our cells, also these cells, if you make your normal animations, you can also run just the animations than the prompts. And lastly, the Load Settings and then you are ready to go. If you click Run all, you can see right here, if you have luck or not, I already have the Colab Pro for me. It's no problem. I'm now connected and I now pay who hours per hour of render time. If you are lucky, you can do it for free. And if you are not lucky, you have to make your subscription right now. So now we only have to wait. We need to give permission to our Google Drive as soon as this is all done. And then we get our animation. As soon as this appears, just gave normal permission. Take your Google account and accept. And now our first frame start to render. And I must say this takes a bit of time. We are right now of four from 200. And you'll see also the Zoom in action. We zoom in right now and the Zoom in, my mind is maybe a little bit strong, but that's totally fine. That video is only, for example, and I will see you as soon as all 200 frames are rendered. Take with me or less, look at our motion parameters. So the angle we rotate clockwise and counterclockwise a bit faster, then again clockwise and even faster. And the last frames are static. The Zoom first, it is fast, then it's that thick. Then we have a slow Zoom and Zoom will go backwards. And the translation x will do really, really funny stuff. Let's just see how all of these will look in a video that's 200 frames long and that will be made in 12 frames per second. And there we have it. Now, let's just see what we do. First, we should zoom and behave in waves. And then it should be pretty ugly relatively soon. As soon as we get our other settings. Yeah, just like that. X camera movement is not really nice. But look at the cats. The cats are relatively good. If we just delete the camera movement of the x-axis, we will have a pretty nice video. I think. We've started with an apple go to banana, standard coconut. Then we start really shaking. And the last things are our cats. The cats have vibrant colors and everything is added is just this camera settings on the x-axis. They are really not optimal. They are way too strong. And of course you can also save your image just pressed for a point right here, and save it to your computer and it will be automatically downloaded. But you don't have to do that because that is all also in your drive. Just go into your Google Drive, you know that path. So AI, stable diffusion, this right here. And we have made the folder banana for our outputs. So in the folder banana, there are all our 200 frames. And if you want that, blend them together or let's say in After Effects or in blender, you can totally do that. And if you go down to the last frame, so that's our last frame. You will also have this text document and your video right here. So you have your video and all your frames already in google Drive. And by pressing on the three dots, you can also download it directly on your computer. And of course you can also download these frames or IT, or every time you wanna do that. And in the next video, I will show you a cool trick that you can do with this frames in your Google Drive. And that's just the beginning. We can make whatever we want. We can animate this however we want. We can start with our own big tours, we can start with our own videos. We can make whatever we possibly can think of. The only limitation is your own imagination. You can make add very thing in this tool. And in the next lectures, we will go any further and we make animations with our pictures, with our videos, we will take the video input and so on. After the course, you will know everything. So see you in the next videos. 30. Letting people age and the resume animation: Maybe you saw some videos how a person gets older on the forum diffusion video. In this lecture, we will take a look how we could possibly do that, because that's really, really easy. We start with the basics. So the first thing we wanna do is Chose to the, in this example, I want to make this animation 100 frames because I be honest here, I want to save some time. We don't need The angry or any rotation. So we put zero here. We don't want to have any Zoom. We want to have a static picture. So we also delete all of this book, no Zoom right here. And the camera behaves moment of all, access should also be zero. So we simply delete all of this and both are, so here are quick zero translation y of course, also at 0.3, the things don't matter right now because we used to the video, the strength. We can increase that a little bit. Let's just roll with 75 right here. Color coherence. Let's play with HSV for this example. We don't need any video input. Their resume animation, that is one thing that I want to show you right after this. The prompts, of course we need to change them. So we start at frame zero. Let's just say with a beautiful woman, 20 years old, high-resolution trending on art station. We can delete that for this example, we booked a little negative prompt. So the mind is minus neck, ugly, blurry, limps. Let's just make this a close up. So we start at frame zero. Beautiful clothes. Alpha woman, 20 years old, high-resolution enter negative prompt is ugly, blurry and extra limps. Let's just say extra. We can simply copy this prompt and both data right here in every prompt cell that we want. Let's just make this until here. Are a woman should be 30 years old, right here, 40 years here. And let's just say they should be 50 years old right here. We will not change also the frames. So 24 days, 44, this, 64, this. And let's just also made D right here and make a woman also 60 years old. And you can play with that how you want they killed. Also be a little bit older or let's say something like this. You can play with these things. The load settings are fine. They see it doesn't matter in this example, scale, It's all fine. Maybe we lean a bit heavier into our prompts and there's one that batch name. Let's just say woman, safe behavior. You can use that fixed seed for this, but I roll with either because people change if they get older. So either works fine. In this case, we don't use any image settings. The last thing is the video input. So we make just 100 frame video to add frames per seconds. Worked fine for me. And we uncheck this box. So that video really I've created immediately and we simply run, all right now, or if you have all installed and your drive is connected, you can simply run chest this box right here, animations, after that, prompts and after that the lobes and things. But in my case, I need to run all this right here is our animation. So our a woman starts at 20 years old and she got older until she is 70 years old. And we have not the prettiest results. I think here in the middle ground there resolution is not the best. That can be because of this seat or it can also be because of the prompts. There are two solutions that you can take right now. Maybe. I think this is because of the strength right here. So if you want to make here better outputs, you can either decrease the strength of it. So our strength with zeros 75 and maybe stable diffusion leans too heavy into this strength. So it makes a bit ugly results. So we can simply decrease the strength debate. Let's just say 0.65, 40.60. That's totally fine for this right here. And you can also do the following. Let's take our example, but let's assume that we really like our first frames. And let's assume our last frames did something specialist, something that we don't like. Let's just make our family thing right now. Let's say we don't want to have a woman on the last frames, but it should be a man. In the age of 55. The woman decided that it wants to be a man right now, are beautiful close up of a man, 70 years old, high resolution I also included this shot with our canon. So we now have also included the camera, the rest of the frames, let's assume, let's just assume we really liked them and how we can do if we wanna make only the last eight different new. That's a really simple. We go into our Google Drive, AI, stable diffusion. We go into our woman example. And now we need to find the frame that we don't like. So in our example, we can start that frame 78 or 79, or even AT if we want, let's assume we want to start at frame 79. What we can do possibly only to start with this frame, just click on it. Copy this until the underscore here before the other stuff, we simply copy this. We go back into Colab and we use this right here. Resume animation. First, we checked this box and we put our frame right here. And now we simply do the following. I decrease the strength of it, and I used this resume. So play in the settings, playing the prompts. So we'll remember until frame 78, we assumed we had a good output. Play in the prompts. And now play the load side things. And we will restart our animation path from foramen 78 or from frame 79. If you think your output is good and DataFrame AT just do that. But I think our output is not the best. So I uncheck this box and I, and I wanna run all prompts you because our animation is not perfect. Now we get our final output. I think it's a bit better right now. The frames are more clear. I think. The last thing is demand. You can see if we take a camera like we did with our man, the output gets even better. Our prompts right now are really simple or not good, but the prompt chest with they can get more detail. You'll see that first prompt, they are not optimal, they are not perfect. But just with the add of the, of the camera, we could make better outputs. And we can even change this woman into a man with a simple prompt. And remember, if you like your first frame, just use your resume animation and you will start your animation from this specific frame that you built here if you check this box. And we should also write better prompts. That is just for the example, we can simply copy prompts from Leonardo or from lexica. And I'm sure we can get better outputs than this. But for this example, I think it's perfect. You'll see what happens if you start with not-so-good prompts. And the last prompt is a bit better because of the camera, you'll see the output Israeli pattern if we use a camera, for example. And that was it. You can make people older with some prompts and some tweaks. You don't like your anti or output. Just run all, just tweak some examples. Just tweak a bit of parameters. That is how you work in the forum. If you don't like a specific number of frame, Let's say you like everything until frame 50. Find frame 50. Copy the number posted under the resume animation, check the box and run it again. And you will start from frame 50, 60, or the specific frame that you pull tear. In the next videos, we will get further and we use our own pictures, our own videos to start our animations. So stay tuned. 31. Your learning experience: Your learning experience. I recommend you to learn in the forum and to learn you have to use it. So you should make your own animation right now. Try to type in what do you want to see? You can make a static picture. So put zero in every camera movement. Describe your prompts just booked in what you want to see and play with the seat and the strength. Play yourselves and just let me know what you got. I'm sure you got something cool. 32. Init Image: start with your own foto as keyframe: Start your animations with your own big tours. I already told you that that's possible. You can start your animations with your own videos and with your own big charts. And in this lecture we will take a look how you can start your animation with a picture of view. First, of course, you need a picture of you. So let's just take a screenshot. I think I have found a really natural one. So the next thing is you need to upload this into your Google Drive. Simply go into your Drive in AI and make a new folder right here. So let's just make this Screenshot folder. For example. In this screenshot folder, you need to upload your pictures. And in our case, it's this picture. As soon as the picture is uploaded, you need to go back into Google Colab. Now, we animate this thing. So what we will not make AI tools here to the video, I want to make it, let's say 120 frames. We will. Now we don't want to have any angle, so we don't want to have our rotation for this one. I actually want to have a little bit of zoom. Let's just say we start with Zoom that goes in. But then after frame, let's say after frame 60, we want to have a little bit of a negative Zoom. So that is the right this, we don't want to have any movement on the camera access free. This stuff is not important right now. That strength, I roll with 0.65. We don't need any of the other stuff, but right now we need to make our prompts. So we start at frame zero with what prompt we want to start. We use some frame from lexica. First of all, I will not be a man with a beard because my beard is not pretty and it's not that big. So copy prompt back right here. At frame zero, I want to turn same plate into a man we've appeared. After that. I want to transform into, let's say, into a zombie with this right here, I will not be a zombie with some flowers. So copy prompt, and let's put it right here. And the last thing I want to be is this right here, so I will not completely die and be as Collette. The other prompts, we activate them with our hashtag, and now we have to choose our frames. So at frame zero, I start with this portrayal of a man that is totally fine for me. We have a zooming until frame 60, I think. And after that we start to zoom out until frame 55. I wanna be the zombie and the metals Collette, I won't be after frame AT as soon as we start really to zoom out, I want to transform into this scale that and I think that's it right here. The load side things, we need to take a resolution that works with my picture. I have the picture in 16 to nine, So we need to take a resolution that works with that. If you struggled with the resolution, just take a look right here. And I think this right here works good for me. So 854 times 480. So we don't render all the day. So this right here is 16 to nine. That is okay. That C doesn't matter. We'll leave this seat. Steps 50. Does Kayla want to put it back down at the default settings of seven? This right here is all okay. The batch name drugs for me we want to have are either seat and this right here, this is now embarked in it. We need to check this box marked enter strength right here. This tells you what the first frame does. The first frame of our new animation in this case, would replicate only ten per cent of my picture. But I want to increase that to 90 per cent. That gift smoother transitions. So my picture is of course that the first frame, and after that, we increase it until 0.9. So that first starting frame of our animation is 90 per cent similar to the picture that we get. Stable diffusion. This image, that's also important. We need to connect our picture to this. So we give the path, what do we need to do is simply click here on this folder. We need to connect our drive. So we press this right here. After that, this cell will appear and you'll need to click play on that gave permission. Now you'll derive showed up B are here and if it doesn't, just press Refresh right here. So now we need to give the path. So Dr. My Drive AI screenshot and that is our picture. So press on it Copy path. And we want to take this path and bolted right here as our innate image. And that's basically it sold. That first frame will be replicated in 0.9 of the strength of my picture. And after that, this strength right here will be used. So this 0.65 that we already put in, our brands are okay and everything else works just fine. So now it's time to play. So play the sad things, played a prompt, played a load settings, and now you have it. It's, your output folder is not connected because in the course it takes always a little bit of time. We need to run all again. So simply run and run all. You need to connect everything if you're works normally and don't make your course how you do that? You are already connected. Now that right here is our video. And let's play. You see I transform relatively quick in the man with the beard. Then I get to the demon and to the skillet. And also the zoom behaves exactly how it should. But if you are not happy with your outputs, of course you can tweak it. If you start with an init image, maybe you want to have an image that's more similar to that. Now what do you have to do is the following. Just increase your use in it to one. That means stable diffusion will replicate your picture one-to-one. So your picture will be the starting frame. After that, we could also choose to tweak the prompts a little bit or something other. I think I will not rotate a little bit on the last frame. So I animate this from frame AT as soon as we get this kel that we will now start to rotate a bit, frame 82 points. And the rotation should not be that fast or rotation of one is enough for me. Zoom, I wanna make it a little bit stronger and in both directions a little bit stronger. That is how you work with that. So now we have 120 frames, a little bit of a rotation from frame AT, we start to zoom in a little bit faster and to zoom out is also a bit faster. I want to increase the strength, also just a tiny bit, maybe to 75%. I think the prompts, they work guilty in our example. So what we do now is play the animation, play the prompts, and play the load side things because the resolution was fine. Everything worked relatively good. And now you can see the transformation begins again, and I see you as soon as this is finished. So we now make a video of this later video. And I think now we get something called the first frame is exactly me. So stable diffusion chance to replicate at me. And the animation goes like this. The Zoom is a bit faster. We start to rotate and also the backwards Zoom is a little bit faster. I think the animation isn't bad, and it starts with our own picture. It starts with our own face. We transform first into a man with a beard, then into an-an that with flowers. And lastly into a scale that this is how you work with the ballroom. You make something go render. If it's cool, Perfect. If it's not cool, go tweak your parameters and render again. That is how you work here. If you don't like the last frames, go find the range that you don't like, put it on the resume animation and start the animation from this frames. And this is the trick. You just need to try it. And in the next video, we will see how we can start with our own videos. So just the videos that you saw trending on TikTok can then stuff. So see you in the next one. 33. Start with your own videos as starting keyframe: Let's start to animate our own videos. And there are several ways how you can do that, but the first steps are always the same. You need a video for him or herself. So let's start with something like this. Right now. I'm in shortcut and I simply corrupt the video just like I need it. So we have this video right now to start with. My video is relatively short, but that's totally fine. And you'll see, I'm just grabbing the camera. And that right here, That's our last frame. We need that last frame. If you work with shortcut or da Vinci, just simply export your last frame if you work on their own because you can totally do this all on your phone. Just take a screenshot of this last frame. I already did it and that is my screenshot of the last frame. This screenshot we need to upload into our Google Drive. So let's go into AI and we bought our screenshot into the Screenshot folder. The next thing we need to do is, of course, go to Google Colab. And the steps right now are nearly the same as we did with our older video. But our key frame right now is the last frame of our video. Now let's animate this thing. Also. This time we worked in 2D max frames, 120 ROK angle. I want to rotate, but I want to rotate faster in this one. The Zoom first two main and then zoom out. I think that works really fine for us. This stuff right here works all the great thing my mind, strength 65 now to the prompts. Always wanted to be a Greek gods. So that's our first prompt. So copy prompt. We go back. The first thing I want to be is the Greek god. Copy. After that, I have found again as zombie. And the last thing is a full body portraits of random. And I think that frames work fine, but I just want to tweak these things. So from frame 60, I will not be this horrible thing. And the venom. So 0-50, I'm a Greek god. 50-90, I'm a zombie. And 90-120, I'm venom. The other things are deactivated. The resolution works fine with this. All the other things works also fine because we have already safe that last time. So now we just need to give the path of our innate image. And that's basically just the same process. So use in it, of course. Then press this order. Mount your drive again. If it's not already mounted, drive. My derive AI of course, screenshots. This screenshot right here is the screenshot that we need. So Copy path bolted under the init set things. And we are ready to go. So run all our round, just the cells that you need. But in my example, I need to run all again because I am not that fast right here. Now let's see what we got. And you need also to check this right again and are using it. The strength right now is at one and it replicates nearly exactly my photo. But you can also use 0.60, 0.70, 0.8, or in most cases 0.9. That's the thing that I would recommend you to use if you want to animate your own video. But one works, also good for me right now, let's just see what we get. We should first domain then start rotate after that, Zoom Out. Yes, we are the Greek god after that, the zombie. And lastly, we are abandoned. I think the video is really, really cool right now. I need to download this to my computer or you can also of course use it from your drive. Now we take our video and make it into shortcut. We have to put the two things together. Okay, And now I did a really stupid mistake. Just locate it. I took the wrong picture to start the animation. I took the wrong screenshot for our folder, but we need to change this. I will change it right now. The only thing I changed is that strength right here and the strength down. The strength. I want to increase it a little bit because I think that it's better and the strength on the US in it, I also decrease to 0.9. 90% of my first frame will get replicate that. And I simply use now another image. And I hope this time it's the right one. Nice. That's the new video. So we actually start now with 0.9 of the strength. I transform into the Greek god than that zombie. And lastly, the venom we start to rotate, that's fine. So download it and into shortcut. I already added a little bit of music. I'll let, I'll show you that later. As you see now the video is really good I think, or it's better. Let's just see how the translation works. Now that are my last frames. So perfect, I think that's perfect. That translation is really smooth. And I transform into the Greek god really smoothly. I think that works fine for us. One last time I want to see it. After that, I simply export it with the music and you can listen to everything. And as you know, you can tweak everything that you want if you don't like it, just tweak it and don't be such an idiot as me. Use the right screen shot to start your animation. And that's the video that we got with the music 34. Video Input: Lay prompts over videos: Video Input in the forum. You can put your video in the forum and overlay prompts. And that's really, really easy. Just take a look with me. The first thing we need is a video. If you have a video from you, find, if you don't have a video from you, we can also go backslash.com, search for a video that we will now overlay. You can download your videos here for free. I think I found a good video. I want to roll with this video for this animation. So there are two persons that are dancing and I think they need a really good prompt over it. I already downloaded it. You need to upload it into your Google Drive. I recommend to make a folder in the AI folder that these named video inputs. In the video inputs there is the video that I uploaded. I call it Video Input default room. So we don't make any mistakes this time. We go into Google Colab and now we need to animate this, but it's really, really simple. All we have to do to set the animations from 2D to video input. If you choose Video Input here, that said things, all doesn't matter. So we can leave our max frames as they are. And also the emotion parameters will get totally ignored. All we have to do is to give the path. So we need to take our video input and gifted video path, and that's easy. We just click on the sidebar. As always, we need to mount our Google soon as your drive is, that you need to find the path. And as always, Dr. My Drive AI, in this case video inputs, Video Input, the forum copy path. And we need to put our path right here beneath this extract frame. If you put here a one, every single foramen will get x-direction. And every single frame we'll get that prompt over it. If you type too, every second frame, we'll get to extract that, then we'll replaced with a prompt. And you can also take free. You can take whatever you want. For this example, we go with one. I want to have every single frame of this. But if you want to save some render time, just put two right here. All the other things you can leave at the default settings. What's important are the frames. I want that this connects all of these frames. So we built our hashtag gets always and we will not run with the same prompt for every single frame. And I think this cyber punk from lexical should work well. So what Cyberpunk we will not take? Let's just take this one right here. I think that's relatively good. So we have a big prompt and a negative prompt. So copy prompt, put it right here, minus, minus neck. And we also take the negative prompt. And as you see, that's a really, really big prompt, but that's totally fine. There are various enough space for this prompt. Image settings is 16 to nine. And I think our pixels here work well. And I'm not totally sure, but it can be that this bigs let's get totally ignore because our video input already has the resolution. That seat, we can use this seat right here. The c doesn't matter that much in that example, a scale of seven works fine for me. And deceit behavior I want to take either because a little bit of diversity is fine for this video. The batch name, I will not have this right here. Cyber punk dance. But the init settings, they are now important. That's drank is the most important part of this thing. If you take one right here, we will get exactly our video. So stable diffusion will not run any prompts. If you put 0.9, 90% of the original video will be original and only ten per cent will wait into the prompts. I think things between 0.4 and 0.8 work really good. And we want our role with 0.8 for this example just to look what we get. And we want to go down here because this right here is also important. I want to have the same frames as my video animation has. Their video has 25 frames per second. You can simply look at this at the outset things and I think we are ready to go. So now it's time to render all and we will see us as soon as this is rendered. This right here is our video that's in Google Drive I think that strength is a little bit too low and the bronze, the bronze are not perfect. We did the prompt of a 4k barrier, and that looks a little bit silly in my mind. But you get the idea. I think I want to run this again. Let's just take a look at the original video, for example. The original video is this. And you'll see the prompts. They get a polite, but we don't do a great job. We need, we need other prompts and a little bit higher weight. I actually typed this prompt myself are beautiful, highly detailed cyber punk. And so beautiful, colorful masterpiece, award-winning picture, vibrant colors, clouds on the ground, the cyber punk standing on clouds. And I think that prompt will run better. I hope it at least because you need to try this enter strength, I want to increase it a bit in its settings. We need to decrease actually this strength to get more of the prompt. So 0.7 or 0.65 even. Let's run with 0.7 and we see what we got for this example. Just to save some render time, I want not just render every second frame. And let's see what we got with this one. You can also read it right here. So export one every two. We export every other frame. And that should be faster. Now that is our next video. And I must say, I like this more. As you can see, it's just in my mind a little bit better but also not optimal. And I have to say, I want to run this again and again with other prompts and with a fixed seat. So see you as soon as all it's rendered. And there we have it. There go to old zombies from walking that database shop for me in this example. Just like a set, you need to play with your settings. You need to play with the strength. You need to play with different set. You need to play with different stuff. But I think the video right now does a relatively good job. And I want to save this. So simply say if it onto your computer and just let's take a look at the original video. And you will see the difference is really, really big right now. So that's the original dancing video. I think the video is good, and that is our video. The Bronze data, relatively good job. And you can see all the dancers and the poses are relatively the same. We have found, we have found the right strength. And every thing for this. Let's just see what we did. Of course I took another problem. These are walking dead zombies. I also increased the scale to eight. I took a fixed seat. The strength, We'll see your 0.78. And that's basically it. And we did a video with 25 frame and I x-ray that every frame, not every second frame. So I think the best output we got right now. And of course you can still tweak these. You can still make this how you want. Try different prompts, try different settings, and just experiment with it. You can really do a lot with this link embark, for example, did the whole music video with this technique. They animated their whole music video with stable diffusion 35. The firs 3D video: I want to be the greatest. Everybody on the patient, I look around, I feel like everybody is the fake is somebody guess every day and I'm hoping one day I blow up in a basement. Same in the top is so big and I don't think that I think is amazing. Way for my day when a play and sold our souls were 1,000. Our first 3D animation. And for that, the first thing we need is of course, our video and our key frame to start with. And I want to do some fancy **** with my hoodie. So just a moment. So I think I got something here. I take the last key frame and make a screenshot of this. I already uploaded it in Google. We go back to our example. And now let's animate with age three the right here. We make this animation 350 forams, the angle and the zoom. We don't need this because we work in 3D. We have translation, translation, x translation, that for the Zoom and translation, Really why my goal is basically to zoom to the object and move to the right. But I always want to see my object. Then when we are here, we also rotate a bit back and also zoom a little bit back. I hope this works. I really think it works. You can see here how the camera excess behave. So that's right here is the x-axis, that's the zoom and theorists that 3D rotation way. And in later lessons I'll show you exactly how you can write this and how you can write this really, really fast and easy. But right now, don't worry about this. If you want, you can copy this. I can also link it. What are the next steps to do that color, I think HSV works fine here. I think the noise is fine if we just leave it at 0.02 for every frame, their strength, I want to maybe, yes, I want to roll with this strength and the restaurant theories of fine. The next things are the brands. And now let's see how we find our prompt. I actually found some prompts and some prompts I wrote myself. So from frame zero, I want to be a black metal skeleton rapping. I think that's nice because I have my hoodie after that, that foramen 100, I will not transform into the goals. Try that. And I think that's also nice because first we have the wrappings, Catlin, and after that the goals dry. That is basically nearly the same thing, just with some fire on it. And I had that prompt right here in my midtone account. At frame free handwrite, we want to transform in a zombie from walking dead. That prompt we already had in play. I just like the prompt and I want to see how we got to this. If we transform after the goals try to into a zombie, I don't know how we behave, but we will see. And I actually also added a few frames after frame 350. I will not be a zombie and I just put this brown TIGR from lexica. Our whole animation is now 400 frames long because I increase this right here just a bit to 400. So we have the last 50 frames as another zombie, I want to just transform again. The next thing I did is the seat. We just use this seat. The image side things are 16 to nine shots with a random normal pixel number, not too big so that we don't render all day long. There's kale, I increased it to eight. The batch name is cyber punk dance. We just use the same folder because it makes no difference that the Z behavior, I just write it with a fixed seat, but that doesn't give the results that I liked. So I took ITER because sometimes if you take a fixed seat, stable diffusion can lead to heavy into the prompts and the outputs where we're just ugly. I must admit it. I use are in it image. So I checked this box, the strength, I increased it to 0.9, and I simply put the path to my drive right here. And as you see, I took the right key frame because I named it who the keyframes. So no issues right there. We don't use a mask at all and everything else is ready. So now we go render and as soon as everything is rendered, I will see you again. We have now nearly all of our frames. So we are now at frame for 893 or 400, but the animation stopped. So we need to go to Google and find our frames. Both decks here, right down to the resume. And we did that play again. After that, I created the video So uncheck this box. I took 30 frames per seconds because our video has a lot of frames and I don't want to have a video that's too long. We go into our folder and here are all our frames. And also the last thing right here is the video. I downloaded the video, and let's take a look. What do we do? So the first frame is me, of course, and it's 90 per cent me. But you will see I will change relatively quick into the skillet and ds. And I think that rotation also do. Thus. He's chop. We get into the goals for either into the zombie right now, yes. Nice. We rotate. I think the video is relatively good. So we take this video and we need to put it in short cut to our other video. Now I want to see how the transition is. Nice. I think it's really good. So you'll see I turn immediately into these wrappings, Collette. And I think we did a really, really great job in transforming the first frames. They are all relatively Gotye think. So let's just play the whole video. I think we do a really good job in this one. So, yes, all we need to do is maybe put a little bit of music over this. And I think we are ready to go. Our animation is really, really good. I think maybe we could change the prompts a bit. Maybe the goals try that from the beginning would be the better option, or maybe node zombies at all. But I think we did a relatively good job with this. And you have seen how easy you can create these videos and you can create every thing that you can imagine. There are no limitations. So just be creative. Play with the prompt, play with the settings, play with different stuff. Maybe put a little bit of music under it and you are ready to go. If you start your own animations, do maybe something where you have a movement and out of the movement started randomization. You can make this or you can clap or something, and you take your specific key frame and then you transform it. Or just like I did with my hoodie. And then you start transform yourself in every thing that you want. And there are really no limitations. And over the next videos, you will learn how you can animate this camera movement easier. We can use keyframe, generate the rest. We can use jet GPT. We can just simply use different tools and make our own workflow out of it. So stay tuned and try to animate your own cool videos. I want to be the greatest. Everybody on the patient, I look around, I feel like everybody is the fake is I make every day and I'm hoping one day I blow up in a basement statement at the top is so big and I'm going to shade that I think is amazing way for my day when I play and sold our souls were 1,006 36. ChatGPT as personal assistent: In this video, I will show you how I use GPT as a personal assistant to make my animations. First things first, we need to prime church EBT. So I told him You are a pro when it comes to diffusion models, camera access, angles, and rotations in 2D and 3D helped me to make incredibly good videos in default room diffusion model that creates individual frames that can be made into a video. I will post you some information here. You read it all and answer. Okay. Only first information. I posted simply the first information from one of our Quick Guides. And I told him, but wait, I have more. Next information. I posted all the animation set things from one of the quick guides. More information. 3d animations like things, again, from one of the quick guides. And then I gave Jack GPT an example. So this is an example of settings I like. And I posted simply the 3D rotations and what they did. I asked, Would you say, you can help me write animations in 3D with the forum like in this example. And check GPT basically said, Absolutely, and the more information we give GBT that better he get. So you can go on and train Chet UP te, further in this chat. Or you can simply ask the following. I have as seen in mind, all in 3D, object sits on that chair right in front of the camera. The camera zooms in and then rotates to the right wherever other object appears. And chatty be Degas, it feels practically the following. He knows that it's zooming in is the translation set in 3D. So he does a really good job. And the rotation right is defray the y-axis anti basically just give me all the things that I really need it. So we can simply put this in stable diffusion and I think that should work. Or I know that works because I hear a lot of animations this way with Jesse Beatty. And you can do that too. Just gave chat D the right information and he will provide you the right answers. That is, how you work with jet GPT. You must be specific. You must give chat UP te the right information. After that, you simply need to tell chat GPT, what do you want to see? Another example, Let's make it a bit more complicated. For example, next scene, 400 frames long, gave me only the mask file object in front of the camera. The camera zooms slowly in, goes to the right, but rotates in a manner that the object is still to see in front of the camera. After frame 200, the camera rotates to the left there zoom is faster and our other objects is to see until the last frames. And there we have it and we have even our quick explanation what we do. So of course, we need to select 3D, our frames to 400 angle and zoom with doesn't need it. After that direct translation x, I think we need to tweak this. So the translations that is the Zoom of course, then the rotation for EY -15 and after that plus 15, that might be a bit high, we can or should decrease that maybe to ten. But we could simply copy and paste this stuff into the forum. And let's just see what we get. And if you're not happy, remember, just tweak your outputs. It can be really that easy. And now I also told check GPT that he can or should rotate from frame 220. 20 also did a great job on their rotation that. So let's just try this thing. I called it all into my Google Colab notebook. Now it's time for us. We have all the translations, all a set. We need 300 max frames and we will not simply built the strength, the strength thes. Okay, in this way, I think we can even increase it just a tiny bit. We need to use proper prompts, of course. And like I said in the last video, I want to start with a ghostwriter right here. And I actually want to be the ghost rider. And then the last frames until we really start to change camera angles and start to zoom in. And that is basically frame 200. So the last 200 frames, we wanted to be something else or something else will appear in our animations. We will see what we get. I think this male zombie drinking our coffee should appear. So we basically just copy this right in here. And we change the frame number to 200. The other frames we this active weight with our hashtag. As always, we need to make the right resolution. And I think I want to increase the height just by five pixels because sometimes it's not perfect with, with our format. Maybe it works, maybe it doesn't. We will see this scale is alright, Everything here is fine. Our batch name. That's all Okay. In my mind, I see it works. Also. We need to use a neonate, the strength of 0.9. We need to give the path, again drive my derive AI screenshots. I use the who the key frame because I think that with the who The wasn't that bad is the idea. We built it right here and everything else is fine. So we need to run this thing and we will see if it works or not. And I got some results. I wasn't quite that tab. I tweaked it. I spoke with jet GPD, and after a while I got this and I think that result is pretty good. Just take a look with me. So we started just to goes dry there and we zoom a bit. We rotate, debate with our cameras. And now comes the rotation and the transformation of the prompts. And I think all in all, it's not bad. Again, one last time. And again, we need to tweak things, we need to try things we need to render, render again, run again, speak with GPT. That is how we work here. As you can see, I just spoke a bit with JEPD, so the objects that bears were relatively quick. I wasn't happy I got new outputs and so on. You just need to talk with Jesse Beatty, a combination of trial and error, sketchy PT and keyframe generators, which we'll talk about later, are the key here. You need to try, you need to explore, you need to render, and you need to tweak. That is how we work here in the ballroom. Use GPT, use also maybe the quick guide takes the quick guide if you're not sure how you rotate or how a camera behaves, just look quickly at the quick guide and you will know it. After that, you can tell chat GPT what was wrong, what he can make better, what you can make better. And maybe later how you can adjust this with key-frame generators. And you will make really, really good videos. I'm sure about that. Just explore yourself 37. Use other and specific models: I already told you at the beginning of the course that you can use your own models in stable diffusion. But how you can do that in the ballroom, It's actually simple. The first step that you need to do is to find the specific models. And I see it all the time. Vapor like dance and stuff. They like demons. They liked this things for the animations. So we need to find a dungeon model. Adoption model that I see relatively often is this right here. It's from cv.com. You simply press this button here and download it to your computer. And you'll see the file is relatively big. But if you downloaded it, you need to upload it to your Google Drive. And the same thing on the Hugging Face. You can find a lot of different models on Hugging Face. That's one model that I see that most people use for pictures that make good outputs, for faces, for example, they make relatively realistic faces. With this analog diffusion model, you need just to click on this link and it will start downloading. As soon as you download that all the models that you like, you need to go into your Google Drive. You need to go into the folder AI. You go into them all this folder. Here, you'll simply need to put your model. Right now, I have three models here. So simply upload the models that you like to use for your animations. And it makes no difference what mountains you upload. You can use any model right here. And then you just go back to your Google Colab. And backing Google Colab, you'll need to go to multiple setups. Model checkpoint. You need to choose custom. So simply click on Custom and then everything you have to do is on the cast them checkpoint path. You need to give the path to your model. And as always, Dr. Mydrive, in this case models and the models will appear that you will not take. Simply press again on them all the copy path and that path right here in it. And it's all set and done. If you have a model that makes warriors, for example, really good, you should use them all for warriors. If you wanna create some warriors and you can go on with this, how you like it. And in these models you can also find some prompts. Sometimes. For example, if we go into this tangent model, you can click on it. And in the model you will find a lot of example. And this example is this right here, and that's the prompt right here. And the negative prompt, you can simply copy this thing into your Google Colab and that's basically it. And if you like, you can also copy the seat if you want to have this exact thing and you're ready to go, just simply find the brands that you like. And you can copy them into your Drive. And you can make exactly the stuff that you can see here into these models. And this is how you can use custom models inside of Google Colab. It's really easy. Just find them all others that you'll think that makes the best outputs for your specific needs. Then you need to download them, upload it into your Google Drive, use custom model and then give the path to your model and you're ready. And if you have latch, the guy that has created them all, Let's also some cool prompts for you. So just try it out. 38. Keystring generator for faster workflow with deforum: The website right behind me, that's a key string generator. And with KEY string generators, you can, of course create key strings. But how exactly that work, you will know it after that video. First things first, you need to go to this website. And here you can animate everything that you can possibly imagine. You'll have to type in what key frames do we want to see how long your animation should be there? Animation, for example, start at frame zero right here, and it goes and DataFrame 100, you can increase that. How much do you like maybe until frame 1,000. And here you can see following. We start at minus one and ends at plus two. You can see this right here. This is plus two and this is minus one. And now you can set the values that you need. You know, if you're working on translation x, translation y, and also the translation set for the Zoom in 3D. You can work with values that go from minus ten to plus ten. So if you want to create animations for this free access, it would make sense to take values from minus ten to plus ten and simply gave their key frame generator these values right here. Let's assume you want to create an animation that is 150 frames long. And now you can simply make the animation how you like it. Just think what x is. Do you want to rotate or what access you want to animate? Let's start with translation x. Translation x does this. So what you want to see from frame zero and frame 150 on the x-axis. Let's assume you want to start, of course, at zero. So simply click kids €1. And then we want to, for example, make the animation until plus two-and-a-half. But it should go back, and it should go back even further than it started. So it goes to minus one. You click at minus one. And now you have right here that KEY string for all these animations. And you can go wild with this. If you click, for example, edit control point, you can pull on this and you can animate this, how you like it. For example, we like this thing right here, and that's now our x-axis. And of course you can't just copy this string if you press copy string. And now we put it on our x axis and they are released from frame zero we start, we should start maybe also the animation at zero. But after that, you can see how the thing behaves. It behaves exactly like we told him. Soda, x-axis, short first, move to the right. And it moves to the right in a really funny manner. After that, it moves to the left and it goes even further to the left. Then we have started. And you can do this with every axis. Let's just make an example. The y-axis, I like to hold them still. I don't wanna see anything on the y-axis, but I want to have translations that I want to have their Zoom. And you can also of course, make the Zoom maybe like this. You want to start zooming in at 04:00 A.M. 50. The frame should go from 50 until 150. And you know that translation that so their Zoom, you can also go with values from minus ten to plus ten. So we leave the exact same thing right here. So now let's animate the Zoom. The Zoom starts at frame 50, and we just simply wanna zoom in maybe the highest value, 2.2, 0.8. For example, we start at frame zero and the Zoom should behave like this. We want to have a really, really funny Zoom. You can even see that frames right here. So this right here is frame 50. And at frame 100, for example, we want to have a really funny Zoom. So we zoom in in n. Well then we make a quick zoom out. Then we statically zoom in until we are a bit higher. So now just copy the string and we want to post it under the Zoom. Completely delete the Zoom, and copy the Zoom right here. And remember we should start maybe a zero here. And the first forams we want to also have static. We also animate the first frames right here. So we start with zoom, that Maybe completely static, and that's it. Now on to the next things. We have translation x, translation y. We want to have that translation that we have as Zoom, and it starts at frame 50. Now we have the rotations. And on their rotations we work with smaller values. So it would make sense to change this. I want to animate the whole axis, so 0-150. And the values are like to make them from minus free until plus free because that or the max values with we will now work. And now you have to think a bit. What do you want to see? You'll have your object. And the first thing is our x-axis that do something like this. And from frame 50, we start to zoom in. And zoom behaves also a little bit funny. So it would make sense to animate the rest of the access with things that you want to see. Remember, if you are unsure about the rotations, just take a quick look at the quick guide. That for 3D rotation, x moves or rotates that camera around the x-axis. The rotation y rotates it around the y-axis. And of course, they're free the rotation that Rhodes the camera over. I will not make something like this. Maybe we can start at frame 50 with our 3D rotation x, so that we may be doing the following. We are static. We zoom in and as soon as we start to zoom, we go a little bit upwards. We start, of course it's the euro. We go maybe until frame 60 or something like that. We go static or nearly static, maybe even a bit down. But after the 60, I want to rotate the camera upwards. It should rotate upwards, maybe a value of 1.6. I think we want to rotate upwards relatively quick. And I want to also animate this just a bit. I want to have funny thing right here. It should behave like this. So copy string and just booked it at our rotation for e, the x. The next thing that we want to see is how we want to behave on their rotation free the way. Remember that 3D rotation y rotates like this, and I think we should rotate also a little bit in that direction in our example. And we should take this also. Maybe we could even take this same settings right here and also put them under our translation. Why? I wanted to see how we behave if we simply put this also right here. So in theory, we should do something like this. Our object is that then we are static first and go just a little bit with the axis. And as soon as we zoom, we go upwards and to the right. Rotation radiates that. So basically also their rotation in 3D, we want that chest. Keep that. Now to the next side things. So the max frame at 150 because our animation is 150 frames long, three the angle and Zoom, no values, that's all fine. The 3D rotation, of course we need to delete this thing. We don't need any fancy perspective flips because that's also for 2D. We don't want to have that strength and noise is fine. I think maybe we decrease the strength little bit, but the noise is fine. Coherence with lab is also fine. We don't need any video input. I also don't want us start with our picture of me. Write this basic example. We will not just make normal things, our prompts. So we want to start off course at frame zero. We start with our beautiful colorful donkey, vibrant color masterpiece trending on art station. And it should transform into an extremely disturbing horror film photography of walking dead zombies. So again, I just want to play with zombies. And why I chose frame 70. I chose friends 70 because at frame 50 we start to zoom in. And from frame 60 on, we rotate with our camera. Maybe this disturbing zombies can appear behind the donkey, or we have just to see how this warm behaves. Image settings are completely fine. The seed makes no difference. Scale and steps is find the Euler sampler is also fine. We will not call this donkey. The seat is either. That is fine for me. We don't use our name on it. And I think all the rest is fine. So now simply go render. So I click Run, run all and I see you as soon as this all is rendered. Right now I'm in Blender and let's just see what we got. So while our donkey first is nearly static, we start to move to the right. The Zoom hits in. And the other Bronze Star start to change. And we rotate to the right and upwards in the 3D motion. So just like we told him, but I'm not really happy with the outputs. We need to rotate maybe a little bit slower. We need to maybe activate. Notice Zoom is fine. Maybe be killed. Start with the movement of the x-axis a little bit sooner but slower. And we need to simply fine tune this fine-tune performance. If you want to, fine tune acts as if you want fine tune everything that you want to fine tune. I will skip that we think in this video because that is up to you right now. You know how to create everything, how to use KEY string generators. And if you get an output, you just simply need to tweak it. Tweak it until you've got something cool 39. Audio keyframe string generator synchronize music and animation: Audio, two keyframes, string generator. This thing blew my mind as I explored it the first time. There is the possibility to use a specific music or a specific soundtracks and to extract the key frames from specific parts of this music. For example, just download the music, which of course, you will need to have their permission to use. You need to use copyright-free music, maybe from the YouTube audio. Take for example. You can download it and you can upload it in a door where you extract a specific part of that song. For example, I already downloaded the song and you can go to Lao Ai. Ai is my favorite tool for extracting these things. And it works really easy, but you have to make your subscription if you want to do more here, there is a possibility to test this thing. And I think you can test it for about 10 min. And after 10 min, you need to make your subscription. If you don't want to make any subscriptions at all, you can use phone Nick mind, and on phone Nick mind. It's basically the same thing. You just upload your music and then you can extract what do you want now here, let's just make an example here on Lalla AI. But it works in every tool nearly the same. You can also run local things on your computer. The first thing you need to do is, of course, upload your song. So we simply press select new file. We can drag and drop our file right here. As soon as this is upload that it starts, then you need to select what you want to extract. And in most cases that rums work best. So let's extract the ramps for this specific example and we'll make it don't refer to seconds. So that is not too big. As soon as everything is done, we need to go to our audio to create forams during generator. And here, of course, we need to upload our file. We can drag and drop our files of only the drums right here in it. And you'll see it. We have 34 s of our song, and we have here our keyframes. All keyframes are here listed. The first thing you need to check is the frames per second. You need to know how many frames you want to create. For example, if you take 24 frames per second right here, you will get an animation that has 800 frames. If you take 12 frames per second, animation is of course shorter. It is only 408 frames because the soundtrack is 34 s. And if your time is 34 by 12, you get the 408 frames. If you time it per 20 for you, of course you've got a lot more frames. The next thing right here, that's the function. We work our example with the zoom and the default settings works perfectly well for the Zoom. But what does this formula mean? Let's just make a quick example. So we have one plus x to the power of four. What does this mean? For example, if we go into the forum and let's take the Zoom for an example, the Zoom is a multiplier. Of course. This right here would be a static. Zooms on, Zoom at all. And you can multiply it by this value and you will zoom slowly in if you work on other things like maybe the y-axis or the x-axis. But I like to zoom the most. You need to change this. You could change this to only x to the power of four, because x is always that ramp and to the power of four. But because Zoom is our multiplier, we need to work with this one plus x to the power of four. So we simply multiply this, and every time that RAM hits, we will get a zoom. You can see it right here. For example, at frame 403, we don't have any zone 404. The we don't have any Zoom, but at 405, we have our Zoom. And that will be exactly when the drum hits. And if we work with our Zoom, of course, you can also change this x to the power of two. And you maybe think that x to the power of two makes smaller adjustments in the zoo, but that is not right in this formula, smaller is bigger. And I tried to explain to you why. In our example, x is smaller than one. And if you multiply x with something and x is smaller than one, a higher value to multiply means a lower number. Let's just make a quick example. So that right here is our example. If the dram doesn't, it said all we have zero If the drum hits really strong, we have 0.3. If the drum hits just a little bit, we have 0.1 and our formula will go like this. One plus 0.3 to the power of four equals 1.081. And if you take that same formula, but with a lower multiplicate, or we get the following. One plus 0.3 to the power of two equals to 1.09. So as you can see, the value of the Zoom is higher. If the multiplicate there are this thing right here is lower. So you just need to remember, if you take the power of two, your output will be stronger than to the power of four. Because we work with values that are smaller than one. That was basically the explanation. Let's assume we want to have a strong hit. So we work with to the power of two. And because we work in the Zoom, we can leave the rest the same cell chests ended up. And you can see it also right here. For example, on frame 405, we have our Zoom of 1.15. If we take it to the power of four, our frame 405 is 1.02. So a smaller value. If we take e to the power of free, of course, it will get right in between. And I think I want to work with the power of free in this basic example, because I think that's who makes the most sense for me. And now we are ready. We have uploaded our, so we choose a frame rate of 12 frames per second. We chose this formula because we work in the Zoom. The Zoom is a multiplier and we took it to the power of four. We have our 408 frames and we simply press Copy String. Now we go into our collab. We booked it right here into the Zoom, and you can see we are basically ready to go. I will not work with our 2D animation right here. So we simply choose 2D. We take our frames, our max frames are 408 frames, as you know. Border replicate works good for me. The angle we put that zero, and of course we put also the other translations at zero because I don't want to have any movement at all for the translation x, translation y. Let's make something finally arrived here. We want to rotate to the whole video, but the Zoom hits only when also the armpits. The other x's will get ignored because we use 2D animation. Noise and strength is right with this. The next thing we need are the prompts. And I want to simply delete this prompt or this equity weighted. And I want to have the same prompt for every frame of our animation. Just did ramp should animate this thing. Our prompt is an epic battle ground safety, basically in a cyber punk style. And of course, I just copied this from lexica. Our image settings are fine because in the course I think this resolution works fine. Scale is okay, the steps are okay. We don't deal is of course an image, so we don't need this. The next thing that is important to me is this right here, the frames per second for our video, we need to choose the same rate for frames that we had in our keyframe generator. If you take here 12, you also need to make the animation with 12 frames, or it doesn't work at all. If you just take one frame more right here, the whole animation not work, it will work, but it is not at the exact same time as the ramp hits. So now the only thing we have to do is to render all the max frames at 500 right here work fine for me because the other frames don't get animate that. So run, run all and we are ready to go. So everything worked well and I just downloaded the file and uploaded it into shortcut. I took also the audio file. And let's just see what we get. I wanna play the whole song right now. And we're really nice. Our drums and our animation are perfect. It works really, really good. And it will also relatively simple to create. We just took our audio file, we accelerated their drums, then we had to upload it into our keyframes string generator. We need to choose the right perameters, so forth, the zoom, it's a multiplier and we need to take one plus x to the power of 23 or four. It depends a bit how strong you want to have your resume. You need to take a lower number if you want to have a higher value at the zoom, we need to take the right frame rate and then we are ready to go. Just download your animation uploaded into a video editor and just synchronize it with your audio file. And you can see it also right here. The audio file is exactly the same length as our animation. So it worked really, really well. No need to tweak this thing because I am happy. I'm sure you can also make really good animations that are synchronized with your audio files. So good luck with that. 40. Complete Projekt: We've seen a lot of different stuff and let's just use this stuff. Let's just use this knowledge to create a full project. I have something in mind and I already started something. Just take a look with me. I sit here in my kitchen and I just try to drink your coffee and no, Normally, I don't wear glasses. That is just for the video. I nearly didn't see anything with this glasses. I put on my glasses, I drink their coffee. I see something in their papers, and I do write this days right here. That is our last frame. I want to start our animation from this foramen. Maybe I could include something right here. Something like AI is going to take over their warmth or something. And as soon as I see it, I spit out my coffee. And from here on, we want us to art my animation. So we take a screenshot of this picture and we'll load it up in Google in our life, you already know how to do that. Let's go in Google Colab and we want to make our animation. Our animation should be in 3D. And I think the right forams or something like 300. I think that is more than enough for our video like this. And maybe I even can take this video for the intro or promo for this course. We will see how this goes. The next thing we want to animate is all of this angle. Of course, I pulled us the Euro because we don't want to rotate at all. The zoom, we don't need the Zoom. You already know if we want to zoom in 3D, we need to animate that translation set. At the first axis that I want to animate is the translation x. And now we need to think what we want to see. You know, we start with our key frame like this. And we need to start to animate our prompts. I think I want to get into a zombie as always. And maybe after that we tweak how operon. So we need to take a look what we want to see. So the first thing is, of course, our access, how they should behave. And I think I want to animate my axis like this. Right here is our object. And I want to start the Zoom in translation, x and translation. Why does nothing but we start to zoom in and then the 3D rotations rotate up and to the right. I think we should change prompts as soon as we start to rotate up. I think that should look relatively nice. And I wanna take KEY string generators to make my animations faster. And just for the sake of this tutorial, I will also animate the X axis, but I think that wouldn't be necessary in our example. But just let's do it. We can always tweak if we don't like it. So the first thing we want to animate is the x-axis. I need my three hundreds of frames from minus ten until plus ten. And let's go about the x axis. I think they should be relatively even until the frame 150 at 150 starts our magic. And the x-axis should go to the right, maybe just a little bit. So something like that. And I think that is fine. Just copy string. And we put this at our x-axis translation. Why I don't need that at all. And tear I wanna start exactly at zero. That translation that, that's something we need. The next thing I want to animate is the Zoom. I think I want to have a Zoom that starts really slowly and increases a bit. So we start at frame zero. We go, I think DataFrame 150. After frame 150, I want to increase the zoom drastically, maybe 23. And it should happen on DataFrames 160. And after that, the Zoom should slowly be much, much slower until the end of our animation. So that right here is our Zoom. So copy the string and under translation set. The next thing are the 3D rotations. I want to rotate to the right at frame 150. So we go back right here. This time we need lower values. So minus free until plus free. Because you know, this axis behave much, much more. I want to also have, of course, the 300 frames. We start at zero and we should go maybe until frame 150 with zero rotation right here. But after that, we start to go upwards and maybe Until a value of 1.55 ink. That is fair. Or let's just play with one right here. Copy string. And at our x-axis, same thing I wanna do with our y-axis. And now the only thing that we need is our rotation. And I think I want to have a rotation or we can even let this deal and maybe re-run it if you, if we think we need a rotation afterwards, the next thing we need to take is that noise. I think we can start with a noise of 0.2. But at frame 150, some magic happens. So we want to increase this after frame 150 a bit, maybe to 0.0 for their strengths. I think that strength is right for us. We could work with age as v In this example, I think that makes sometimes the better outputs. We need no video input and all. The rest of this is fine. And now let's go to the prompts. In what we want to transform. I found a prompt on Leonardo and I want to transform in this. I hope that it works. So just copy prompt. We go back into call the app. We put it right here at frame zero. We also have a negative prompt, so minus minus neck. We need to copy our negative prompt and we built it right here. So that is our positive prompt and we want to change prompt as soon as we rotate. Remember we start to rotate upwards and to the right at frame 150. And also the Zoom starts to be more quick. So I want to change prompts may be at frame 160 or 170. So that may be new things up a year as soon as we zoom out of our frames. So I want to activate this prompt right here, maybe from frame 100 and let's say 165. Start with, we need to tweak if it's not good. And what we wanna see right here in the background of us, in the background, something like heme should appear. So copy prompt. So at frame 165, we want that paste is writing. And at my first prompt I want to add right here, blade girl on that arc from drinking our coffee because I spit it out my coffee. And basically that's it. Notes that things are fine. Our resolution works well. The seat doesn't really matter, but we have a seat. Euler, 50 steps, we need to lean a bit heavier into our prompt. Maybe eight. Brown said things are fine. We will not name this. Spitting my coffee. Use the CT are either seat I think works also well. We need to use an init. So we activate the innate strength is 0.9. That is fine. And now we need to connect our Google Drive and gave the path to our image. We do that as always. We click right here, we connect our Google Play. This cell gave permission. If it doesn't appear, we need to refresh. We go to derive my derive AI. Now we need to take the right screenshot, start keyframe. So cup Copy path. We need to book the bath right here. And we can go on What's next. We don't use any mask. We don't use this things at all. But I want to have 24 frames because my camera or social two with 24 or 25 frames. And I think we are ready to go. The max frames down here are to 500. I think all the rest is relatively fine. Yes, the prompts change right here. And the only thing that we need to do right now is go render. So simply run all or just run here the animations after that, ran the prompts, and now run the load side things and we are ready to go. And I see you as soon as everything is rendered. And here we have our video. Let's just take a look at transform. The transformation is good and smooth. We start to Zoom. I think we should rotate right now to the right. Also this prompt right here, they are really nice. I think now we change prompts and rotate to the right. Yes, Change prompts to the right. I think now more noise should start kicking. Yes, more noise so we can create this stuff faster. And I think the whole animation is a relatively good. And of course we can and probably should tweak that if we want. Maybe their camera or rotations are a bit too harsh at the ending and a bit too slow at the beginning. And we could simply tweak that if we like. But I think I don't want to tweak this. I think that is relatively good, so I downloaded it. We upload it in our shortcut. I tried to put it right here on my video. And I think that transformation should also be relatively smooth. Let's just take a look. Now I split my coffee and now I should transform. And you'll see the transformation is really, really smooth. And then I'm relatively quickly, this lady. Yes, I think the video is relative guilt. Let's take a look at the whole video. First with my stupid glasses and can't see anything. But that's okay. Start to split and now I transform. The ladies are illegal. The Zoom is also okay, it's not too fast. I think that's relatively guilt also the rotation. I think I liked that. It's maybe a bit too fast at the end. But I want to simply leave this as it is. We can always tweak it if we want. But I wanna leave it as it is. I think that's really, really nice. And you can see thanks to the increased noise at the ending, we can create the stuff faster. So stable diffusion keeps up with creating our new stuff. We can also make this longer or other or animate this further. We can tweak and do whatever we like with this. Now we got to a full project and I think, you know what you have to do now, Exactly. Start your own projects. And it's a relatively simple, just make a video. And it doesn't matter what we do. Start with animation, something like a cliff or something like that. And then start your animation. Take your last keyframe, upload it in your Google. Gift the path and use the innate tweak, the prompts, tweak the camera access, use keyframe generators and you can and will create relatively good stuff. We did exactly what we want to do. Our object was right in front of us. We zoomed. After that, we rotate upwards and into the right corner. And the zoom was faster and we rotate even faster. We did exactly what we want to do. And thanks to this course, you can also do exactly what you want to do. If we want to tweak this, maybe we could decrease the speed a little bit of the last frames. We could also change the prompts are little tiny bit later, and the outputs are really amazing. But I think also this output is relatively good. We should probably just overlay a little bit of music over this. We are ready to go. So let's go create your own animation. 41. What you shold create and where you can get inspiration: You may ask yourself, would you should create with the forum diffusion? And I really don't know it because you can create everything with the forum. You can create to the animation, 3D animations. You can overlay your prompts over your video's, over your own videos, Overstock videos over everything. Dan, you can make your own animations. You can start with a video from you with a video of an animal, with a picture of an animal. And you can also animate this in to the 3D, every single thing, how you want to see it. So just simply go and explore the forum. And if you really don't know what you should make, go to TikTok and Instagram. Just find some information. Go to TikTok and look videos just like that. That's maybe the most popular video on TikTok and the video that really, really got viral. I'm sure you'll find a lot of stuff that you can explore. You could also go to Instagram, maybe to their site of stable diffusion. They have also some cool videos. As you can see here. They have animated a cool free the video right here. And they have also done something, I think with oval lame prompt right in this video. And you can make every single thing that you want to see. Just go on social media and explore a bit and maybe try to rebuild it, make it better. Give it your own style, give it other prompts. Try to animate the camera rotations child's IQ. So on TikTok, I'm sure you'll need a lot of time if you want to recreate thing that you see on Instagram. If you want to recreate maybe a video like this, I'm sure we'll need a lot of time because there are a lot of details in this video. And if you maybe go to his profile page, you will see that the camera rotations of his new or videos, or maybe most of the time this same because it is rarely, rarely, rarely time expensive to animate your camera like this. But thanks to this course, you'll know how cool you can do this a lot faster. You can use chat GPT, you can use key stream generators, but you need to try it. You need to explore it. You'll need to tweak it. As soon as you have something that you really like, you can use it for more animations. Simply copy your Colab notebook and do it again over and over again. You can also create your whole music video, extract maybe the rams or something other, and create your own music video. You can make this how long you want. You can even do hours and hours of these animations. So go explore yourself 42. Learning experience in 3D: If you made it until here, congratulations. Now I have a learning experience for you. And I think that learning experience, wealth, they care a little bit to do. I think you should create your own video. Make a video from yourself. Do something like drinking our coffee. And then start with an animation or something like that. And start your own animation. You should make that animation in 3D. And you should rotate the cameras how you like them to rotate, you should use keyframe stream generators to make the process smoother. And if you want, you should even use your own models. Just think about what you want to create. Do you wanna create photo-realistic, mildest yields down our log diffusion model? Do you want to make the immense? You could also take our demand model or you can search for your own model yourself, load everything up in Google. So the last key frame of your video and them all gave the right path and start to animate as soon as everything is animated. Tweak it until you're really, really like it. Use GPT if you think you can't do it alone. But I think with the key string generators, you are really, really good to go. After that. Load everything up in your favorite editor and underlaid with a bit of music. And now you should publish it on Instagram, on TikTok, on Twitter, or maybe on YouTube shorts. Just choose what you like and remember, takes the right resolution for your platform, Instagram and TikTok, you don't use 16 to nine, but if you want to make a YouTube video, you should use 16 to nine. So just remember to use the right image dimensions. As soon as everything is done, as everything is published, hit me up with a message. You can also send me the thing. Just let me know if you can create cool stuff and I'm sure you can create it. So have fun exploring the forum. And if you need help, I'm here for you. 43. Kaiber a solied and easy alternative: You may ask yourself if there are alternatives to deform diffusion. And yes, there are alternatives. And some of them are executable, good. Others of them are, in my mind, just too pricey and the outputs are not that good. One alternative that is really, really easy to use. And it's not that pricey Khyber. If you go to Cairo or you can go to the gallery and see what other people did with Khyber. But I want to show you what I did with Khyber. So we go to my videos. I've simply created two videos because I must admit it that was for free. I don't want to pay for this because the control is not that guilt. The interface is really, really easy to use. You simply upload a picture of you and you can type in this prompt and you can type what do you want to see? So camera, for example, zoom in. And you can also include a rotation if you want, and you'll see the output. So the output is relatively good, but you don't have a lot of control, and you can only do this in 2D. So in my mind, it's really not worth it right now. But the other feature is relatively good. You can also overlay your pictures with prompts. As you can see right here, I simply uploaded a video from me where I am. But he basically just humbling around in my room. And you'll see what I did do it with overlaid prompt. A man dancing in the style of illustration, highly detailed, simple, smooth, and so on. And it was really, really easy and it is intuitive. You can just upload it. The prompts are newly created for themselves because you just have to type in two or three words. The rest is the prompt magic from Khyber. And you get your video at the end. And you can also animate whole videos. I think Lincoln Park did their music video completely with Khyber. So in Khyber or you can work, it can be a bit expensive if you need to do a lot of it, but you have not the best control. A my mind, you should work with divorce him diffusion. And there's also wrap diffusion. Rep, diffusion is another Colab notebook. And in my experience, rapid diffusion makes good outputs, but it can only overall a prompts over your videos, but you can use control net and stuff. So rep diffusion is relatively nice. But I want to dive into it because I'm not an expert in red diffusion, rep, diffusion is easier to use than the forum, but I think also you have more control in the forum and you can make cooler stuff with the forum. No other thing you can make such good 3D creations. After that, there is also one last app, and the last step I wanna show you in the next video. 44. Loopsie and disco or wrab diffusion: Loop CAI, it's an alternative, it works, but it doesn't work that goat, you have nearly no control over it. But you can use it on your phone. And it's easy if you want to create some animations with loops, They just try it. Log yourself in, pay for the bucks a month, and have no control over your outputs. I wouldn't recommend you to use loops, but I just wanna say that loops there is also an alternative. So for me, it's not really a good alternative, but you can use it if you want. But this right here, this could work diffusion. So that rapid diffusion that I already told you in the last video. We've wrapped diffusion. You can overlay your prompts over your videos. And that works also in Google Colab. But I must admit that I am no expert in rapid diffusion. If you will not drive rapid diffusion, you could maybe use this free protocol that works also, but the best and newest version you have to pay Alex on patriarchy. The most things are relatively the same as in stable diffusion DeVoe room. But not everything is the same. But I think if you try just a little bit, you can also create your first animations here with rap diffusion. And just like I said, you can overlay your prompts over your videos with rapid diffusion. And you can do that really, really good with really a lot of control. But my preferred way is the boron diffusion because of their 3D rotations and the animations that start and go to nowhere. So I think the forum is the way to go 45. Stable Diffusion [Wrapfusion] in Google Colab: Stable Be fusion. That's also a cool tool that I will show you how to use in this course, step one, of course, you know it, you need a file. We do basically the same stuff as soon as the frame comes around where we want to transform. I cut the video and then I extract the part where I want to be transformed afterwards. Of course, I go one step back, because we need that later. Now I have the file. I put it simply into a folder because we need that file later. The next step is we need to go to Patrim. We work in a Cop notebook, but the Cop notebook we will get only access if we subscribe to someone on Patrin. He give always the New West versions and he gives tips and so on if you have any problems. There's also a discord server for me. I think this is really nice. Here I am into Patrim. The name of the guy is X L E L A or most likely XS. If you read it from behind, he is really active. He gives some updates every few days. And right now we are in version 0.20 That's only a preview. We need to be a power member. If we want to have this, I'm not a power member, I just spend that $1 But we can have stable B fusion, the version 19.8 You need to download your version that you like. Sometimes the newest versions don't work that well. A version that I always liked using was the version 0.16 You can also simply press on this file. If we press on this file, you can download it on your computer. The next step is to click on the Cop notebook. I will link you everything. This right here is a normal notebook. You need to go to uploads and you need to find your file. Right now I use version 0.16 because this works always really, really nice. We upload here our version and then we are ready to go. Now this cop notebook maybe looks a bit overwhelming because it is really, really big. There's code included and so on. But this is really easy to use. We don't have to do a lot. First step, I like to go to run time. We need to make sure that we use GPU, and I like to use a stronger GPU change run time. The version 100 U works fine. Also the 100 U works fine. This works for me. If you use four U, this will take forever. Then we scroll down to set up. We need to press play on the first cell. Of course, we need to give permission because this will get access to our Google Drive. So we give permission right here. Then we need to scroll down a bit. We need to go to basic settings. What's the batch name? We just call this test. But we've do these because I think I already have a test then. What's the width and the height? Simply the resolution. My video is in 16 to nine. This resolution works fine for me. Now we need to go to Video Input Settings, and of course, we need to give our path. We click right here as always. Then we can upload our video right here, or we can even upload it in our drive. But this right here works fine for me because I just want to make this video. I don't want to have this in my drive. So this is totally fine for me if it gets deleted afterwards. As soon as this is uploaded, I called it wrap me, we press the three dots, we press copy path. We delete this path right here and we put our path here in this field. This field is the next important field, extract frame. If we put one right here, every frame will get reworked. Every frame will get extracted and reworked with our prompts afterwards. If you put two right here, video will rendered in the half amount of the time, but I want to have every frame rendered in this video. Then we scroll down if you like, you can store every frame in your Google Drive, but you don't have to do that. But I'd like to do this if we want to work in Blenda afterwards, for example, After that you can scroll down even further until you get to this code here. It's also really easy. Everything you have to do is instead of CPU, we use CPU in this field. Here is model path and control net models. These are simply the default models that we have right here. I want to make the short and easy. So we use the default models, but you can also go to Cvti.com if you need to have a specific look. You can download models for free right here on CVI. Upload them into your Google Drive, and then you can give the path to the model that you uploaded. But like I said, I want to make the short and easy. We use the default models, that's basically it. If you scroll down a bit to non GUY, it comes right here. You can also delete the prompt and use different prompts. But right now, I want to use the default prompt, because we can change the prompt afterwards. Everything we need to do right now is simply run and run. All now the Colab Notebook will do its stuff, but this takes a bit of time. If you scroll up to the top, you will see that every cell will get automatically activated. The co lap Notebook does simply stuff. It installs different stuff, it download stuff. This takes a while. This took about 5 minutes. If we scroll down, you can see the program is doing his stuff right here. So it extracted some frames. Then at some point, you will see a preview under the cell. Do the run. This is the cell that is nearly the last cell in this Cop notebook. If you like this output, perfect. If you don't like this output, you can press Stop right here. And then you can scroll up until you see here, the prompting box. Until you can put whatever you like in the prompting box. You simply include the things that you want to see in this prompt. And in the negative prompt, you want to include what you don't want to see. Text, naked, nude logo, cropped to heads and so on. This works all really good. I think this cyberbunk look, I really like this. I want to transform in such a cyberpunk, the only thing that I want to change is I don't want to be a woman, I want to be a man. Remember, you can always copy prompt. You can ask Chachi BD for stable diffusion prompts and so on. Now with press play again. Down here we have another prompt. Now the only thing that we need to do is play again right here. And then we should transform into a cyberpunk man, instead a cyberpunk woman and perfect. Now this works really, really fine for me. As you can see, now we turn into the cyberpunk man. You can also see this takes a lot of time. I think this will take maybe up to an hour because this is not accurate. Sometimes you can really wait a lot. These are right now, 142 frames, just from experience. Sometimes you need to wait for an hour, you need to drink a coffee. You can see right here all the images they will get stored into my drive, content drive drive, stable B Fusion images out. I go into my drive, now I go into AI stable B Fusion images out the test order. Here you can see the frames come. Now we have now our third or forward frame. And the frames will be collected in this folder. But we can also make a video as soon as all of this is done. I see you. As soon as the video is rendered, just make a coffee. Now we have it. This is right now, the last frame. Now we want to create our video from our frames. All the frames are now also collected in our Google Drive. We can also use the frames and put them in blender, for example. But I like the easy way I want to create the video inside this Colab notebook. That is really easy. All you need to put right here is the right frames per second. Remember, if you have a video with audio, it's important that you use the same frames per second in this video right now. Because if you use a higher or lower number of frames, the audio will not be synchronized. My video has 25 frames, and for that reason, we need to put 25 also right here. Then we are basically ready to render. We click Create Video right here, and then we go. But I already showed you how you can do that. Simply press Play and you will get your video. In this video, I will show it in Blender. We need to go into our drive, We go into our folder where all the frames are. I want to delete the first frame, and then I download all of these frames. These are 141 frames. If we scroll down, I already downloaded them and I saved them into a folder. Remember, you don't have to do that, but I just want to show this in blender because I think maybe it's valuable for some people. Then we go into blender, we need to make sure that we go to the video sequencer and it all should look like this. Then we aim for a resolution that is perfect for us, 1,900.25080 is perfect. The frame rate, I set it to 25 frames, just like I told you, and the output should be a video. Now, be press Add, and we search for our frames. We press image or image sequence. We need to find them. So these are our images. And if we press A, all of them get selected. Simply press A on your keyboard and then press Add Image strip. Now they are basically all added into your blender. Then we need to make sure that we save them in appropriate way so you can leave them into this folder. Or you can create a new folder right here or search for a folder on your desktop. Just do how you like it, maybe we do it into Paps, accept it. Then we need to adjust our frames. Right here. Frames start, one is okay, but the end is at 141. The last thing we want to do is into the left corner press, render and render animation here you can also see the process. If you like, I will see you as soon as this is rendered. This is normally relatively fast. Now it's rendered, everything will get automatically into the folder that we set the folder wraps. Just remember, you can also do this directly in the Covid Notebook. Just press play. But I think for some people it's maybe cool also to no blender. The next thing is really easy, we go back into shortcut. We include our video right here, and then we just throw it on our timeline, directly over our video. And it's basically done. You can see, as soon as I clap right here, the next frame steps in, and I will get into this cool animated stuff. Let's take one less look at this. I think this is really good. Perfect. Now we want to export this, and basically it's done. In this video, we took a look at Stable B Fusion. It's a Colab Notebook. Behind the Colab Notebook works Stable diffusion, it's relatively easy to use. Just use a video clip where you want to overlay your prompts. As soon as you have found your video clip, you need to go into the Colab notebook, but you only get access if you download it from X on Patri. As soon as the notebook is opened, just connect it, press a few play buttons. If you really like use different models from CBTI, then you simply press Run. All of course give the paths and so on. Just like I told you, this will need a bit of time. Maybe you need something like an hour or so to render this. After it's rendered, you can make a video of it directly into the Cole Notebook. But in this video, I wanted to show you that it's also possible in blender, you can download all your frames and you can make your frames into a video. Also in blender, then we just throw it into shortcut. And it's basically done, so relatively easy, but you need to do a few steps. You need to go to Patrian. It's a bit of work, but thanks to this tutorial, it's easy. Remember, as soon as it starts to render your first frame, you can always post this and adjust your prompts. You can put anything that you like into this prompt. You can also transform, for example, into a turtle or whatever you like. Just try it out. 46. What is next: Create and share your animations: What's next? What you should do with the information from this course. I will tell you, you should explore the forum. And after you explored it, you should create your animations. After you created your animations, you should share them on TikTok, on Instagram, or you can even create your own YouTube channel. This video is have the potential to get clicks, to get likes, and they have the potential to generate followers for you. So make videos and simply post them. If you have followers, you can do a lot of stuff. You can sell them things. You can show them things. And it's simply cool to have a lot of followers. Make your own AIR. Start with the volume diffusion and share it on Instagram and TikTok. And I'm sure you will have a lot of fun doing it. 47. Thank you: The last thing I want to tell you right here in this course is, thank you. Thank you that you put the time into watching this course. Your time is valuable. So I appreciate your time. And if there are any questions at any time of the course, please feel free to ask. I answer super quick. I answer every time as soon as I can. One last thing, if you will, can give me a super good writing. I would really appreciate it. And the last thing I want to say is go explore yourself. I think AI is the future and you are on the forefront of the future because you watched this course. Good luck. I hope I see you in the next one.