Master Generative AI using Comfy UI and Flux , AI Art

1. Introduction: Have you ever wanted to create a stunning digital art, T cards, images, videos without needing any kind of advanced design skills, then this course is for you. Maybe you have tried Photoshop, Illustrator, CoraloKvaPcreate, but the process is too complex and it's really time consuming, right? Or perhaps you are an artist, designer. Or an entrepreneur looking to monetize your creativity, selling stickers on ETS or on Rad bubble. Well, what if I told you that you can create with just one click all these images and videos for yourself and for your business and even influencer that is totally AI generative. A with the power of this tool known as CofiUI and it is totally free within your computer. Welcome to CofUINFlux Advance Generative AIP digital artists. In this course, I always teach you how to install Compui understanding the workflow, understanding the interface, generate stunning digital arts and stickers. That is high quality, high quality vector graphics, and even air influencers. And many more things within this CompuI tool. Hi, I am current Acadi specialist with ten years of experience in graphic design. I worked with corporate plans, freelance as a content creator for big brands to help them streamline the creative process. When I first started exploring AI generated art, I realized many people struggling to get high quality images. Either AI images look messy or it was too difficult to get the final look. That's when I discovered the PUI. Now I'm here to teach you how you can use I like A P. It is a project based course where you will learn how to create your own workflow. Use that workflow to generate high quality images, high quality vector graphics, high quality stickers. That are ready to use or sell a project that you can use a control net to control the outlook of your image, training your own Laura to create a consistent character. By the end of this course, you have your own port for you that you can sell, socase and use for your own brand. This course is for digital artist, designer freelancer and entrepreneurs, content creators and marketers with no prior AI design experience needed, I just need your curiosity to learn this amazing tool. So it is for everyone from beginner to advance. This course gets regularly updated. You can share your project with me in the project gallery. I will always give you a feedback and share tip. Are you ready to level up your AIR skills and start creating a high quality digital asset? Let's dive in see in the course. 2. How to install comfyUI: First time going to teach you how to install Comp UI. So go to Google search for Compi Install. That's all you have to type. Look for the CUI the most powerful click it. I will direct you to the Github page, and you can download it by scrolling down and direct Download button you can see here. Click it, it will automatically start downloading for you, and now you have to download I have already downloaded it, so I don't have to download it again, but I will show you the step what it will look like and how to install it correctly. So first thing you have to understand and you must have is NVDA graphic card. If you don't have NVDA graphic card, then it will run on your CPU, which will take way more time and you will lose all the interest how to use ComfUI you will lose all the interest. And it is a ZIP 75, so you have to extract it before you run into your so first, you have to download it from Zip seven, download for your machine. After downloading it, you have to extract that file. I will show you how we can do it. As you can see on your skin, the file will look like this. This is the file that we have downloaded from that folder. It is a 1.5 GB file, and you have to right click, but before you have to install seven Zip, like I have told you, so right click seven Zip, Open archive or extract files or extract here. I will choose Extract to Cf portable NVDA, so it will generate a folder for you, and it will automatically extract all the file in that folder. So Double click that folder and you only have to run NVDA GPU. Double click. Okay. So whenever you Double click at the first click, you will see some window command window will open up and all the information it will download first. So it will take around one to 2 hours to download it. You must have a fast Internet connection, then it will save you so much time. This process is for first time only. So after that, it will open a direct window for you. That is a Come don't have to worry about this thing. It is like a cluster or a web so you don't have to worry and scared out of it. Don't worry. I'm here to explain every part of it. At first, there will be no CUI interface. So what you have to do is that load. Load default, you have to select load default, and it will automatically load a default node system for you. And you can see load, checkpoint amp latent image, clip text, clip encode, K sampler, VAE decode and save. Right now we are using only a default system for you. First, I have to explain you everything from sketch. After that, you will understand all the nodes of what is checkpoint, what is flux, what is clip, what is as amply, what is DAV, all the things I will explain you on this particular course. Obviously, I will teach you how to prompt. So right now, we are going to simply type a simple prompt hours Garden and beautifuple standing. You have to click prompt, so it will take hardly seconds for you to work. So Comfy has generated a basic image for us. Let's change the width of our image. It is right now 152 X 152, which is very low quality. And right now, I'm going to use 102-41-0204. Okay. And prompt again. As you can see, the image it has generated using the basic tool we have right 3. Important Information to Note about first section: Hey, everyone. Welcome to the course. I'm so glad you are here to learn creating art with Kofi UI. In this first section, you all need to do is just watch. This part is to help you get comfortable with Kofi UI tools and see how things work. Just relax and watch it like a movie. No need to do any hands on work yet. This first section takes about 30 minutes to watch, sit back and get familiar with everything. When we get to second section, that's when you will start trying things out yourself and make your own creative AI art. So let's start by watching this first part together. 4. How use Comfy Ui: So we are getting started with the CPUI. So in the pious lecture, we have learned how to install the Comp UI. So in this video, we are going to use the CPUI how we can use it. And we will talk about the important notes that Copy has and their work. As we have downloaded CompuI before, we have all these files that CPI has already installed. What happened next time when you open the CFUI? So next time when you open the CPUI, there are two options. First, run CPU or run NVD. As I've already mentioned in this course, that you must have NVDA graphic card because it will make your work way more easier. So you just have to double click the run NVD GPU, double click it, and the command prompt will open for you. And this time it will not take so much time. It will check the basic information about the machine we are using, and it will automatically open your default browser with a URL, something like this. And it is working locally on your machine. Have already learned to install CompuI and come to this window. Now it's time to make your ideas into reality. So how did I get this window? As you have already installed CompuI using this prompt, you only have to double click this again, and Cf I will open a new window for you in your default browser. It could be Google Chrome or it could be Windows. So it's working 100% locally, no Internet is used right now because there is no web address you can see in your search bar, and there will be two type of window you can see in your first attempt to open the CofiUI. First, you will see a blank screen, or you will see a load default screen. If you don't have any skin here, you simply have to click a load default menu here. In the load default menu, fui will open a beautiful scenery nature grass portal landscape. Pure Galaxy portal open for you. Just click the Q prom. Cofi will work automatically for is the first image that you have generated right now. This is your first image, the CpUIO the creative that are using cofuI have created a wonderful image using cofuI Whenever you click again, it will again generate an image. But this time it will be different because of the seed nature. We have seed here, randomized document, randomized fix. If you make it fixed, it will again generate same image for you again and again. You only have to make it randomized if you want different results every time. 5. Create Basic workflow and understanding the node system within ComfyUI: Getting back to the blank Canvas and right now, just click Clear. Okay. And now we have, again, a blank Canvas and we don't know what to do. You only have to double click. There is a window pop up or you only have to click sampler, or you have to search K sampler here. Come for UI core. You click it, and there is a node has been generated. Now you can see it is kind of a brain, or you can see it is kind of a heart of a compuI node system. You only have to generate model, double click Load image, say Lod checkpoint. It will work like a model. You only have to click to the model menu. It will connect to the model. Now you have to double click again, clip encoder. This will be the text to add to our UI. Double click again. Or you can simply Control C and control V, paste it bring it out here or you can bring it anywhere in the canvas. It's up to you simply in the condition, click on connect it to the negative menu. If you want to resize a node, you only have to drag it like this. And if you want to delete it, right, click the move. If you are mistakenly delete your node, you only have to control the set. I will back again. And now there is another way to add a node. You just have to right click Add node sampling sampler. You only have to search from this vast variety of nodes. Or if you want to add another way, this is my favorite way or it is actually easy to add, you just have to drag out this node and release it in the blank canvas. It will show you the option which node you want to connect with this node system. So you only have to add ampitlatent image here. It will be automatically connected for you and add another node for you. For our heart, we have already had brain which will act which we call checkpoint or model and blood brain eyes, which our machine will see through our text, and it is a latent image. Later on, we will use it as a reference image or we can add a reference image to give our prompt or K sampler to understand which type of images we want or which type of reference images we want to connect with checkpoint and give us a desired result. So right now on the other side of Ks empler we have a Latin, drag it out and you can see VAE decode. There is another output for the VAE decode. Let's first let minimize this sampler. You just only have to click this dot, and it will automatically minimize or it will automatically will minimize its window or if you want to open it, again, you only have to click dot Again, this time we have another image, so it will be a image of preview image. So this time we will see a preview image here. So let's add a preview image, same preview image. So right now it is a complete workflow. You only have to add text. Let's see if we can get an image. Let's make it portal and OT For the positive product, I have added portal, and for the negative watermark. Let's Q prompt. Let's see if we have any request here. This is important for you to understand if we have a problem in our workflow, the CUI will automatically dett that problem and tell you that we have to add another node to complete the workflow. Or if there is any problem, it will tell you right away what is the problem. So you only have to read what is the problem and you have to rectify this problem. Prompt output fall very decent clip encoder. So we have to add a clip text encoder required input is missing. So right now you have to check the color here. We have a VAE color matching with the VAE decode here and all the nodes that have other colors right now turns into gray. So we have to connect it with the VAE encoder and again, the problem we have faced here right now, I have rectified one problem here and another problem has popped out. So we have two problem required missing input. Clip. So we have to check the clip and check where it will be added and it will be at here. And here. So let's first prompt again. Congratulations, everyone. So now you have Master tap resize this image 210-24-1024. Q prompt. Let's click it and drag it out. Now you can see the clarity in this image, and it has dimension of 1024 pixels to 1024 pixels, which is square, and you now can post it into social media. It is a complete workflow you have generated right now. Press Control and hold it by clicking left click and drag it out on the canvas. You can select all the nodes here or delete it. Right now, I'm not going to delete it by scrolling up and scrolling down or click the scroll button. You can drag the Canvas according to you. You only have to click Control click, Control click, Control click, Control click. The interface is really, really simple. And if you know how to use a basic Photoshop, you can understand how we can use the node. Or otherwise, if you have not used Photoshop before, there will be no problem because your friend here will tell you everything about compi. So you don't have to go anywhere else. You only have to watch course. And one more request, please post your first project image with me. I know you are very creative, and I want you to generate a first image, very first image that you have generated using CFI, and post it on project or if you want to share with me directly on Instagram, there as well. It will motivate me to make more videos for you. Now let's generate Lion. Baby cute ion. Cue prom. Let's see what we have here. Here is a lion cute line we have generated here, press control, select these three nodes, Control plusECpy and Control plus Sift plus V paste it here. Wherever you cursor has gone, it will automatically paste here, and automatically, the nodes will connect it. If you only press Control and V without SIFT, it will be paste your notes. The three nodes which we have selected will be paste without connection. So why we have done this, you will get to understand better as we use CPI 6. Comfy Ui interface and group generation: Now if you want to move them, you can move by pressing Sift key, and you can move them as a group. Now you can create them as a group also select by pressing Control, select your node, right click anywhere on the canvas, and it is really irritating for the chrome users. If you don't want this thing to come up every time you click, you should use your browser default browser, which is our Internet Explorer or Microsoft Explorer. So by right clicking that, you need to add group. Group name, group A sample. Now you have created a group here. As you can see, I have created a group here. Now whenever you move the group, all the nodes within the group will be placed along with the group. So you don't have to move them one by one right now. Now let's come to another part. Now let's add a latent image here, like we have added empty latent image for the case Empler we have added. For this case empler we need a latent image, and the latent image will be, let's double click here and add a latent image up scale, upscale latent, add it here, and now add this scale here and ten with here. Now we have added upscale latent image here, and now we are going to upscale it, and we're going to upscale all the way to 1024 to 1024. That's okay it. And one last thing we have to add is reduce the noise to 0.2 because whenever it upscale our image, it will create another variation for us in that image, but we don't want that to happen. So we simply need to reduce the denoise variation here generating first image. Whenever we upscale our image, it will not change our image very much. Just a minor twig, it will do for us after reducing the denoise. Now try to queue our prompt again here. After first generation, now it's working on our image on our latent image. Let's compare both the image that we have generated by using Jagannautbn safenu as C 1.5. Okay. 7. Save nodes as template: You want to save this node as a template. You just need to select this node by pressing Control and select this. Right click, right click on the Canvas. You need to select Save select it as template name, image, scale. And now you want to use that template, you just simply need to select right click and go to Node Template Image scale. We have ed it. This here is our template. 8. Canvas interface change: Now as you can see, we don't have space around here and how to move our canvas. You simply need to hold space and move your curson. Now you can move your canvas anywhere around, leave your space button, and our cursor will automatically work as usual. Now you press space, press and hold space button, and now you can see it is really easy interface you can learn. So you don't have to find a space if your canvas is full of notes. You simply press to hold space. Es ps. You can see it looks really messy, so you only have to click this setting here, you can change your dark mode to light mode. And you can change edit attention. You can change grab, ink render mode to linear or spiral and go to straight. Now you can see all the lines have been changed to State, and it look really nice and easy to use. Milk white, you can see it looks very nice. 9. Comfy UI Manager: Let's talk about Comp UR manager. Whenever you install CompuR in the previous version, there is no manager here install automatically. You have to install it manually before. Now it will be automatically installed a Compi manager for you in all the upcoming update in the latest version of CompuI. So you need to select manager and you can see which node and which node we want to select and deselect or which node we want to install automatically in within the CompuI. It will help you manage all the nodes here. Whenever you install or use a workflow, it will show up pink style of node, which means the node is not available or node is not installed in your system. You need to install them using manager, and how you can install it, you simply need to install missing custom nodes. You click here and it will show you all the results which are shown in the different color or in that bright pink color in the mpiUI canvas. And all the nodes that are not installed, it will be appear here, and you need to select them and install them manually or it will show you that install all the nodes here or here anywhere in the window, you need to install them by selecting all the nodes here. And if there is any workflow that you have installed in your CopiUI, it has missing node and the missing nodes have not appeared here, then you have to find that node in the Github or hugging pace website. Like I have told you, whenever you try to queue prompt, there is error message here, and you have to copy that message and paste in Google, and it will try to redirect you to that missing node. Once you get the experience of installing first node, you will automatically understand how to install other nodes that are not available in CPU. Or we can say, these are extension which you have to install in the CPUI. In the upcoming lectures, I will show them up. After installing those nodes that are missing in the CPUI, you need to restart your CPI, and it is very easy. In the managers section, you need to restart your CPUI that's all, and the new window will be pop up. Now let's get into another part. 10. Install Custome Node: Let's get to another part, select manager and install custom node. And you will see there are thousands of nodes are available within the CFUI. Right now, there are 1404 custom notes available within the CFUI. Let's try to install one of the nodes, try to install it and restart required, and you need to simply restart the CompUI. And if you want to install more than one note, you node here, you simply select them and install them instantly and restart. 11. Workflow in Comfy UI: You can change your custom colors here, right click tighter mode colors and give it a right color. And you can see that color has been highlighted here. You can give your own color. Let's give it a green color. You can change the shape here around box, default, card, default colors, custom colors. You can select any color you want. Hey. This pink color will show up whenever there is a missing node whenever you try to install or use a work flow created by someone else expend click Colors All right. Colors help me identify where I have to type green color to our text broom and it helps me to differentiate speed up my workflow where to find my notes, how to differentiate my notes. Because these are really, really simple notes right now. And whenever we have a cluster of notes, we need to identify them as a group and also as colors, and it will help you in your future projects. You want to use a radimi template from other creator, you can find them from Open AIRt or from Cit AI. You just need to load select workflow that has been saved by other people, and you can use them by selecting and create open. This is a workflow that you can see has been created by someone else, and you can use this and you can use those workflow for yourself. 12. Recent Update in comfy UI: There is a recent update in the Comp UI. It will give you update regularly. And there is a navigation menu you can see here, zoom in, Zoom out, reset view, select space, togal link visibility to remove your node system. There is minor update here, so we don't have to worry about anything. I just have to show you the updates. 13. Load Check points and Trigger words: So let's go to civitai.com. It is important to know that this website is actually safe for you to use. And whenever you open this website, you will see many images with Check Point, Laura, with presets. There's so many preset Laura and checkpoints are available in the CivitI website. Cute at I can see here, and there is a prompt you can see in the center of the picture, a small asteroid is running towards a giant game tamer. They get TD rendering Q towards her. Now, this is a prompt that someone had given and uploaded it on the website. Okay, now we have a workflow. What type of notes they have, which type of guidance they have used, which types of how many steps they have used, sampler and STS information also. To get this exact same result, you need all this information for the CFUI. First, let's understand what is workflow. This is our workflow. All the information, like I have told you, lip amping K sampler, the brain, VAE decode and image sample. This is our checkpoint, dream shape, SD 1.5 and flux, many checkpoints here, and I'm going to update a new checkpoint if you have seen a red mark here like we have Q prompt, and you can see the green mark here. And this gain mark will come to this and this and this and this and this, and the image will be generated here. Whenever you see a red mark or pink mark, there might be possibility that checkpoint is not available. There might be some error that CoVUI has rectified, and you have to remove that error. Let's change to Dave flux, safe tensor, Check Point. As you can see, red dot pink line, and it has already selected an error that we have our error in our work flow. And we have to remove that error. Let's change to load CheckPoint again. And I'm going to download another checkpoint CofiUI Wildcard fantasy extreme Sd Excel, Checkpoint Wildcard fantasy. We have Laura, extreme detail Laura, and A checkpoint here. Here is the checkpoint. I've opened image and tried to use the checkpoint in our workflow. As you can see, there's a fantasy art in this checkpoint. But if you want to download another checkpoint with your preference, you can find it in here models, Checkpoints fantasy Wizard which is let's open this checkpoint. It is actually a a, and it has base model flux one d, and I'm checking for SDSL highest rated. You have to check for highest rated or most downloaded because these are safe to download and more stable. Let's get to highest rated or most downloadable. Highest rated, select SDSL 1.5 Checkpoint. Let's filter it out. Filter by category. Let's get to Vical clothing objects. It's up to you. Let's download this, Laura, checkpoint. And you have to download this in the CopuIFolder. Go to the folder that you have downloaded CPUIOpen CompuI models, Laura Animate Checkpoint. And CPI model checkpoint, and you have downloaded here. It will download it for you, and all the checkpoint have different sizes like five GV, six GV and it is about two GB of checkpoint. And one thing to remember, it has a triggered word, which should be add in your workflow in the text area. Copy this information, and let's try to create a CUI dog in formation. Okay, so we will use this information. You will get this link in your video description. Let's copy all this information. 14. Learn in depth about some terms: So the information we have copied and the checkpoint already have downloaded in the folder, like we have specified in the Cf UI model and CheckPoint folder. Now let's go to our Cf UI window and refresh it. Now let's check. Here is another checkpoint load. You can see now, select it, copy the information, copy it, and these are the trigger word help your CUI software or machine learning to understand that this checkpoint had this particular information and this particular information you have to load using this text prom. Iron Man. Poster fighting Batman. I hope this will generate something amazing. Mm. And let's change the dimension. And again, step 40. Another single figure, you prompt again. Okay, so it actually creating Iron Man, movie poster image and not fighting Batman, but ironman mixed with Batman. It is kind of that let's use image. Let's see directly this prompt in our image and use this negative prompt in our negative. Step 99 and in the negative prompt. In the negative in the positive and step we have given is 99 generate now, Q prompt. It's getting somewhere, not 100% like this, but yes, getting somewhere. I hope you understand how this load checkpoint work. I just simply download it a load checkpoint. Any checkpoint, I have downloaded it, put it in this folder like I have specified. Add positive pn, prompt, negative prompt, and a atentive image, like we have discussed in the previous lectures, there is a K sampler and VA and save image. This is a basic model that CPUI has already in its default. In the upcoming lecture, we are going to expand our horizon and we'll see how all these things works one by one. Right now, we have checked or right now we have load at checkpoint. Is important for you to visit civiti.com website and check for different as how these as work, what type of prom they are using to identify the load checkpoint. Which type of tigger word they are using, create a document so that you will not get lost. Next time you will use Tiggerd Word or prom, and you must have saved this inner document for your future afference. And trust me, it will help you like this one, let's open it again, and they have used flux Dave model, luminous shadow scape neon retro wave by crononit. We will use these type of images, and we will generate these type of advanced images in our upcoming lectures. 15. Ksampler SEED: So in the last episode, we have learned so many things in which we have created an image Las and preset. And in the previous section, we have learned about load checkpoint, clip encoder, text encoder, ampit latent image, asmpler load the ampity latent image and create an image that is visible to us, and the case sampler is the main brain here. So the case ampler will cook together everything and show us a visible image. And it will visible to us after AE decode. In the case ampler we have seed here. This is seed, and seed is the value in which case sampler use the starting point and the ending point of generating the image from load checkpoint or any model. If we keep the seed remains same, if we keep seeds value same, it will always produce the same result every time we cube if we change the value of the seed, it will create a different image. Every time we change the value of the seed. You can randomize the seed by controlling after generate randomized fixed increment or decrement. There are few options you can see in the control panel for the seed value. The minimum value of the seed is zero and the large value is very big. With this large value, you can generate infinite number of images using the same model or same prompt using the same ingredients. Can send the seed number to increment. Whenever you prom, it will generate a image, and it will increase by 16. Exploring Queue Prompt: You can set multiple Q, just like you give multiple command to the printer. You can also give multiple command to Q. If I click it, one, two, three. I have tell him to create three times. It has generated three times. As you can see, it has generated three images for me. If you want to cancel any job, you can see VQ. All the process is running or pending, just simply click Que prompt and you can cancel it out by clicking the cancel button and you can close this window again. You can see the extra options here. Once check and click on AutoQe, it will automatically generate images for you. And it will not stop until you stop it. We have a decreement number here. If you generate image, it will decrease by one. If you keep it randomized, it will generate random number for you, and the probability of getting the same image is actually very, very low because of the large difference in the number. Now let's keep the seed value to fig and generate again. Now, whenever you que prompt after fixing the seed, it will generate the same image every time we click. Now let's change a prompt. Purple I have changed purple to red. Now let's see. Now, it has generated a red color galaxy. It somehow tried to create the same image every time we click the Q prom because we have the same seed number. If we try to change the prom, it will try to create a same image. Every time we try to add the text, but every text has different value in the computer language, or we can say every word has different weightage so it will create image according to that. But somehow it will try to generate the same image every time we que it will generate a different image when we try to change the step to 19 or 20 or Aga value. Let's generate again. You can see there is a difference in the steps. Let's make it 30 steps. It will try to change the images to something similar, but you can see there is a color difference here. There is always a subtle variation in the out 17. Impact of Steps in Result: Keep it in mind. The longer the step, the image will take longer time according to the step. Let's make it five steps. Q prompt. You can see the timing here, only takes seconds or 1 second. 18. Addign Effect to image CFG classifier free guidance : Have to check for CFG value. If you change the CFG value to one, you might not like the results. If you change the value to ten, You might like the result a little bit more. But the CFG value is not directly related to the contrast. As you can see, there is a contrast change in both the image, but you have to check for the CFG value in which it will comfortably create image for you. If you increase the CFG value to 15, let's see what I will get there is a change in the contrast value of the CFG. CFG is not directly change the contrast of the image. I actually give the effect to the image in which contrast is already come to light. So the full form of CFG is classifier free guidance. So if the CFG value is low, the model has the freedom to create images according to the load point or a. It has its own freedom to create images randomly. CFG value is high. The model will stick to the prompt that we have given, and it will try to create strictly or try to follow the rule strictly. 19. How to change all the values in Ksampler automatically using Primitive Node: You right click on the case sampler, it has many options and you have to check for the convert visit to input, and it has many many options here, like C to input, control after generate, step two input. Let's try to convert CFG to input value. You can see the CFG is not here right now, and it has changed its position to the input value of CFG. Let's drag out this node and try to add a node here, and it has many nodes. Let's search for primitive. And this primitive node allow you to add value to the CFG value. You can see just like the seed value, you can change here the control after increment. Let's Q prompt and it will automatically increase the value of the CFG. So just like SD, we can control the CFG value as we want here. It will help you in larger workflow in which you want to experiment prompt with different set. If you want to change the settings of each value each time you generate the image and if you have not get the desired result, you have to change the settings of every option manually, and if you are lazy to change the settings of everything, you can use this type of node, and it will automatically change the setting for you. Let's change our case sampler back to where it was. Now we have to select option, convert input to widget, and we have only one input here. Let's click and CFG comes back to its original place. And we have a node here, which is ampithe, you can delete it out. Let's try to convert another one just to give you an idea, convert visas to input, and you can select cellular denoise, let's select denoise, and you can see the denoise here and add try to add primitive node again. Let's convert it, and you can see the denoise has a primitive node. Now, let's again convert our K sampler to its original place and you can delete it out. 20. Batch Generation Of Images and Naming: Can also experiment with sampler and dular. Most of the time, I use DPM PPM, that is DPM plus plus two and normal caras. You have to expand what work for a specific type of a or lot checkpoint or preset and which model work best with which type of sampler? It's all experiment based and you have infinite number of options. You can check and select what best for you. You can see bad size and you can click one, two, three, and it will generate about three images. Lets Q prompt it. And once all the images are generated, it will show here, let them generate so all the three images has been generated and you can see one s three. It means images are three and you are watching first images. Let's click it. Now you can see the second image and third image. You can switch the images and check how many images have been generated in the one batch and you don't have to queue three times. You only have to click here. Make it one again. You want to watch all the images at once, you can select X. Every image you can see at once. You can change the name of the images, as you can see there is a prefix that is cofuI. So it means it will add a CUI or a number with your images in the name section. So you can change it to bottle. It makes more sense. And if you want to see all the images that have been generated, you can check for the Cofi folder. In the CUI folder, you can check for CfUI output. As you can see, the name is Kofi UI. Ter generation, it will now add portal in your images. So you can see the images that has been generated. These are all the images that we have generated in this course so far. These are some high quality images that I have generated for my course also. I love these images the most. That's why I used for my course, cover 21. Create group and Using group all the information for group: Now let's select multiple nodes. Press Shift click click and you have to hold the shift and click click, click, click. Now we have selected all the nodes here or right click on the case Empler or any node, and you can select the option, convert to group node. Once you click it and you have to name it Okay. Now the nodes we have selected have been merged into one node, and you can see it is a group and it has all the details text area, all the security name, width, height. This is the batch number that we have used, and it is a latent image folder and it's a load model, and seeds and the case sampler details has been updated here, denois and VID code. You can see the bottle here, fixed prefix name that we have updated here. And we have not selected the save image here, but we can add that again, save image. I don't know why it has been removed, but somehow it has been removed. So this is the node that has been generated, and you can select and merge it. That is why I have selected the save image and you can see we have now added a node here, and I have named it text. Now the node we have generated here, you can compare it with a workflow that we have generated before. You can check it a checkpoint, it has text area. It has amp latent image with a bad size, you can see the ampithe latent image details, and it has case sampler details, SIDS control strap and VAE decode, VAE decode that is image has been generated here. Now you can remove that and put here. Now it has a workflow here and I can delete all these nodes, select it, press Control and select the node and delete simply Q prompt. So finally, we have generated an image again. With the new group node that we have generated or we have created, you can manage the group node and you have the power to change the order also. Once you right click, manage group node, and you can see all the node that we have generated here into a group and you can change the position of the node, Checkpoint, Checkpointheckpoint, checkpoint, and save it. Now you can see the checkpoint has been shifted to the bottom of the nodes. It is now convenient for us to see and select. Now it is not looking so much bulky right now. Right click again, manage group node. Now you can set the visibility of the sampler or any of the nodes. Once you select the K sampler and if we don't want the etular or denoise, you can see that we have not used denoise here, once you click it, save it and you can see the denoise option has been removed. It is best for the larger workflow once you create it and if you feel like there is so much distraction, you can remove them, and it's best for you. And if you want all your nodes back again and if you want to see the nodes again, you can convert back to nodes. It look actually cluttered, but at least allow you to watch all the nodes. Let's control. And I have selected this, delete, delete, delete, delete. Why we have not added the same image in the group section? Because once you generate a image, there is a preview image here. Once you add the premium image here, the size of the preview image will not increase, as you can see the increased size here. Now I have converted the nodes back to its original place, and now you can see press control, select all the nodes. Copy and paste here. You can see all the nodes has been now copy it and once you prompt, first workflow will work and after that second workflow will work. Once you select all the nodes, you have to click Add group for selected node. Now we have a selected group here, and now you can change all the nodes position at once and you can name this group also. You can right click on group and AdditGroup. You can name that group and I can name the group into bottle and now you don't have to move all the nodes one by one. If you want to remove the group, you can click group, did group, and remove. 22. How to by pass nodes in comfyui: Now let's double click on Canvas and you can see search node. Let's select for bottle that now we can see the bottle group has been updated in the group section, and now this is the group that we have generated a workflow before. And now we have two workflow right now. One is Above and one is Blu. Now let's prompt. So we have two workflow has been generated, generated its own image without text, and it has worked perfectly fine. And there is another workflow that has been worked again. So two workflow, one cue, both have done their job perfectly well. Now, if we want to make it work, that one of them will work, and one of them will not work. You simply have to select press Control, select, right click on the group section and search for the option bypass group node. Can bypass this group node, and now this workflow is hidden in the eyes of Q prompt, press control that. There is a shortcut to make it bypass select Control M. Now you have hide this workflow, and it will not work again. As you can see, only one workflow has been work. In the group section, you can select, control. Control M simply hide them or if you want to bypass them, you can bypass them by selecting the bypass group. 23. We are now getting into advance section: Can also enable or disable singer note by using the same process and it will help you in your larger workflow. Now let's edit group, remove, select, delete Control and select Control M. I have press Control twice. Once is disabled, and another one is unable. Now with this lecture, we have covered all the basics of ComfUI so that you can understand work with the Cf UI workflow. Now we are going to generate high quality images with different as and preset. After this lecture, everything's going to blow your mind. See you in the next lecture. Thank you for taking this course. I'm your friend Karen. See you in the next lecture. 24. How to update comfyUI and Workflow flies information: Hello, everyone. Thank you for taking this course that you have took this course. Let me tell you one thing about Kofi UI. This Kofi UI or this node system is, I think, kind of universal kind of thing, because if you check for blender and also for DawncRsolve, DawncRsolve is a video editing software and our blender is a three D software. Both these software and upcoming software actually applying this node system in their these node system are also known as workflow. So we are going to create a workflow that use flux. Let me tell you what is flux. Flux is a model in which images are compiled according to your prom, and you will get the visual result of your tech as an image. That's in a simple or layman language, this is known as flux. Model. Check Point. It is not a very difficult thing to do. Once you understand this, you can create your own workflow. Let me tell you one thing. Let's start by clicking the manager. Once you click the manager and click on Update All. Once you update all, all your system get updated because compuI is open source. Develop all over the globe contribute in this machine learning software or we can say in programming. That's why it get updated daily. So I recommend you once in a while you have to update your CUI to get the latest and updated fast results. Once you update it, you have to just click Restart, and it will get restart, reconnecting it, and it will reconnect and another node will open here. Sorry, it's not a node. It is a window or tab. Another tab will open, so let's close it and wait for the will take some time. As you can see, everything is getting updated. Let's jump to the future directly. Meanwhile, it's getting updated. Let me introduce with the workflow that I have generated from the Internet and not the Internet, but from the person who have made intelligent and genius person, I respect all of them, and I have downloaded many workflow and apply in my work. As you can see, these are workflow and many workflow that I love in this the flux or comp need of JCNFleRClick, open. You can see these are node system that flux uses, and you can create your own node system. Or you can save your own node system. 25. ComfyUI latest interface update and walk through: Let me tell you one thing. Every time you open the Comp UI and update it and you will see a new interface, and it is really confusing for you. Like, last time, everything was lower right hand side and right, and now everything has been changed. But it is not that difficult, as you can see. The workflow, you can see, workflow new opened. Wow. It is actually become easier way to use ComfUI and all these. As you can see, there are Q, history Q, and there is library. Wow. You can check for library within the confua right now. Wow. This is actually exciting. You can see a node. Okay, got it. Okay, okay, okay. These are nodes. I thought we can change directly models here, but we can find the nodes here, like search for diffusion model. Diffusion. Yes, can. And it's working perfectly fine. Can we search here? Click is also working. Check Okay, yes, that's what I'm talking about. These are old Wow. Can we check for dreams? Okay. Not bad. We have four checkpoint dreams LV Let's see Checkpoint. Okay, Checkpoint and SDSL one Juggernaut, Lod Checkpoint. Do we have flux? Can we use Flux checkpoint directly? I don't know if it will work because I think this is not, we can check for diffusion model. Is it working? Oh Uh uh uh diffusion model is not working. Okay, so adding diffusion is not working directly. Text encoder, we can check. We can check. No, I don't think it's working fine, but it's working in that way, but I don't think it is working Let's check for unsaved workflow dot. Okay, this is unsaved workflow. Okay, and upscale. Nodes map, upscale. Okay, this is the node map, and upscale is high here. Okay. Sample advance custom node. Okay. As you can see, you can bypass directly. It is very handy. It is actually very handy to use this interface of Cf and I really loved it. Wow. You can check for batch count, like we have discussed earlier. You have to basically experiment all these things so you understand the interface of new CFUI. So let's check okay, white plaque. You can directly change it. Wow. So let me tell you one thing. So let's work on a new workflow, new open, save saves, export, export. 26. Adance workflow in ComfyUI flux overview: Pot of workflow. So these are so many workflow that I have used. So let's check for Image two video workflow. Let me check if that's. This is the workflow that I have worked before. As you can see, there is a prompt and there is a character which is dancing if I can show you that. I think that is not loaded right now. Not an issue, but this is just a workflow for our video. We will deeply discuss all these things in upcoming lecture, so don't worry about that. So these are the images that I have generated, and I have used all these images, four to five images and generate and using them, I have created this kind of really, really, really amazing transition effect on images, as you can see. And these are high quality you can post them online and create a portfolio out of all these workflow. There is another workflow, same workflow that I have shown before. But I have tried to create something, but I have not found it very well. As you can see, this is the one I love. This is not my desired result, but this is I loved it. So what about code image generation will I think copy. Okay, so this is really amazing that we can do anything with this plug. So let me first Q prompt it before I know that one of the node has been replaced by another node, and we have to fix that if that happened. So we have two lets Q and I hope that it work fine. I just jump to the future directly, so you don't have to wait for all these percentage. I simply jump to the future because we have the power of vidalty. This is a prom that already has been implanted. Visa mountain peak down with the Lord Gansia. Lord Gansa and Lord Shiva, I hope you know. If you are outside from India, then probably you maybe don't know. Lord Shiva and Lord Ganesh really respectful God in the Hindu religion, we worship and as you can see, od Ganesha, and Lod Shiva. So it is working perfectly fine. So this is the workflow. I will share this workflow with you. Okay. So you don't have to worry about anything by creating this amazing workflow, you don't have to worry and you don't have to add any extra node and no, no, no node. But I just want you to understand how these node system. 27. Advance Prompt and Diffusion model and Loras and workflow: Let me explain it one by one, and I will share this amazing workflow with you. Don't worry. You can find it in the description anywhere on the platform in the download section. Let me introduce with the flux model loader. It is a flux model der. It is a group. As you can see, there is a group, and we have named this group flux model loader. Load diffusion model. Here is a load diffusion. You can add this by Double click Load diffusion model, Advanced loader, CUI core, as you can see, Lord diffusion model. We have added this, you can see here. Okay, so let's delete. And dual cap loader and it has a load VAE. As you can see, we have downloaded this flux model before, and again, I will explain you how we can download it. Once you click for Flux, you can check for CheckPoint, and as you can see, flux one D. Click on Download. It is actually Hughes file full Model FPP, and I recommend you can download this one. And after downloading this one, this is the image that I have randomly opened in Civit AI. It has Laura and they have shared the number and what does this number mean? This number, how much weight you are using in generating an image. Apart from that, let's check for Laura. Click it and click it. Let's check for it. And once you open it, you can see it has a Laura and I'm downloading this Laura. Do we have a triggered word for this? No, we don't have any triggered word. And I recommend you that you create a document for all your Las and all your flux, all your checkpoint, all your model that you are using all have their own triggered point. Yes, it has a triggered word. You can see that PG style. You have to copy that. If you have not used this triggered word, our model or our CUI will not get an idea how we are going to use this a. Probably the Compi will ignore this Laura. And I don't know how will it react if we don't use this type of trigger word. Most probably it will ignore it. So what you have to do is you have to download this, go to your CompI Windowsp table, CompUI check for models, and you can see that it has a folder name Laura and save it. After saving it, right now, copy this and I'm creating an online dog for this. Yes, the dog we have used before. I'm going to add a triggered word here. And you can save it by triggered word that is triggered words and check and check for any Laura. You have to check for this Laura. Save this No triggered word. And for this Laura, you have a triggered word. And Let's resize it. All my triggered word will get a red mark on them so that I can directly copy them and I can pinpoint them that this is a trigger word. Download another Laura that you can see, download this Laura. It has two files. Download one of them. Same in same folder and it's downloading. Let's jump into the future again. Let's refresh. Refresh node definitions, it refresh update requested, and I hope this will refresh. Yes, the node has been refreshes. Now you can check for Laura. Which type of a that we have downloaded. First RPG, and first is flux realism La and keep it 0.5 weight and so it's two, none, name RPG loa and keep the weight 0.5. Now we have another La. Now let's paste for RPG style, and for this prompt, let's copy this prom directly, and let's see which type of result we get RPG style. And most probably, it has a tigerd word already embedded in this prompt. Let's check for this if we get one RPG. And if not, let's add one then. This is top. We have added Laura SD R style, offset, save tensor, RPG. As you can see, it has already been added, and I don't know if this will work. Maybe I just copy it layered style. I don't know if this work or not. Let's just embed it. Pages. I will explain you about the prompting, in the upcoming lectures, we will talk everything. Right now, I'm just trying to give you an idea how this thing work. Okay? So let's prompt, and I hope this will work promptly. Let's jump to the future again. Okay, so we have not get the desired result. We have a problem with our I don't know, maybe with both of them 0.55. Okay. And 0.5. And let's send that to the 0.99. Let's try again. Okay, there is some change I can see. As you can see, this type of book we can see right now. So what I have changed, let me tell you one thing. Before I have changed a denoise, that is 0.5, and right now, denoise is 0.99. And the weight of the La has been 0.5 and clip weight is 0.5. And model weight is 0.5 Clip weight is 0.5. So I think they both have clashed into each other. There's a conflict between them, and now Hmm. This is amazing. This actually amazing. Let's check for the saved folder. Co PUI, and check for output folder. Come PUI. Check for output folder. As you can see, this is the image that we have generated using flux. The result is pretty amazing. Let's experiment this by not using any Laura. And what what will be the result if we are not using any Laura in that? Let's check it out, Q prompt. Let's jump to the future again. Yes, results are almost same. I think somewhat If you check check if you differentiate in both the images, there is a difference in the texture of these both images. This is kind of pretty smooth type of texture is shown here. And yes, they're pretty much kind of not any kind of realism in them. And I can see there is a realism in them. Yes, you can see the texture difference. Yes. These are imperfect things. Make this image a realistic one, as you can see. And with a somehow I feel like, to be honest, these both are amazing. I really, really appreciate the prom that they have. Given to us. Actually, prompting is not a hit and trial or working with of UI is not a hit and trial thing. It is the thing that once you understand how you have to convey or how you have to explain to this machine learning, you will understand the prompt. That's all. So depict an ancient mystical book resting on a weathered wooden table in a dimsly tilt in a dimly lit, magical setting, the book's edged, cracked leather cover. Feathers, a vivid illustration of a dark and int forest, leather cover features, okay? The forest is bathed in the pale moonlight. It's nulled, leafless trees stretching their roots over mossy forest floor, the winding steam cascade through the scene, forming miniature waterfall that glisting in the sterile light light. Artwork appears seamlessly embedded into the book's leather as if enchanged. Surrounding the book are atmospheric details. A few melted candles in ornate brass holder cast a warm flickering glow, their light illuminating a dedicated silver crescent moon, amulet and a tarnish curtain nearby, a quill and ink wells sit alongside a scrolls and weather Tom. They are feathered gold, lettering, hinting and the Sanchis of knowledge. The table is rich with textures. Is surface is scattered with dust and fainted scratches. As you can see, the prom that depict this is somehow actually depict all the prompt that we have given to the Cf UI in the background, soft focus, Silet of other magical artifacts and book hint at the scholar of Maze workspace. The color plate balance more greens and brown with a warm golden glow of candle light and a silvery scene of moonlight. The scene exclude mystery and timeless enchantment, inviting viewers to imagine the secre hidden within the tombs page. And this is art style that we have given. 28. Difference in Flux Dev and Schnell and Commercial use license: We are getting deep into flux and understanding it better. I have gave you the overview for the flux, how we can use it and about the workflow. I will explain you briefly about the flux, a new model that I absolutely love. I will walk you through how to install and use a different version like the Dave and Snell version, which one works best in terms of quality, speed, and setting. So let's understand what is flux. What is flux? Flux is created by Black Forest Lab, a team of smart researchers and engineers who used to work with stability AI. Now they have introduced the flux version, one model family, which performs really well, but heads up this model needs a good computer for smooth performance. There are three versions Persion day version, snail version. Pro version, API only, not downloadable, Day version, best quality, but resource heavy, Snail version, fastest, but slightly lower quality. The license for these versions are a bit different. Dave version, no commercial use, you can use the image it generated for commercial purposes like selling them, but not to train a computing model. It simply means you just can't use flux to train another model so that it will compete with the flux. Nail version fully allows commercial use for both model and its output. But for the snail version, they provide all the rights to the user. It's always a good idea to check the license files yourself for the most accurate details. On the left hand side, they have shared flux pro in the middle, flux Dave and flux Nail on the right. Precision in text rendering, both flux pro excelt accurately reproducing text within images, make it idle for design requiring label words or phrases. Whether it signals book covers or branded content, flux pro delivers clear and correct text integration. However, Flux Dave does a decent job. Snell structures to render text effectively. Complex composition. All flux model demonstrate exceptional skills in understanding and executing complex composition. Whether you are dreaming up elaborate fantasy, realism or precise product visualization, Flux effortlessly, bring your multi element prom to life with stunning accuracy. As you can see, here is the difference. Anatomical accuracy. All flux model demonstrate great anatomical accuracy, especially in the rendering of human features. These models consistently outperform previous open source alternative like stable diffusion three, SDxl and so on. In creating realistic and proportionate body parts, significantly enhancing the quality of character focused image. As you can see, here is the difference. It's actually look pure realistic, but it is a fake. It's look good, but as you can see, the difference in hands, here and here. And now here. I looked like a filter has been applied on this image, but look realistic. Fine tuning color. All models excel at accuracy reproducing specified color palette, making it a go to tool for maintaining brand consistency, evoking specific moods or bringing creative vision to life with exactness, whether crafting marketing materials or pursuing artistic endeavors. Flux model delivers the desired color scheme with unpalett precision. Okay. Yes. Here's a difference. Same prompt, three output. I like this one, by the way. But as you can see, it is more realistic one. So alters and aesthetic. All three flux motor produce turning high resolution images that capture intricate details delivering top notch clarity and aesthetic. This is a real one. It looks like someone has clicked this image and look about this Dave and about Snell. It looks good, and it look good, too. But this one is the realistic and aesthetic look. Yes. Here's the difference with these three models. As you can see, these three models have the difference or whole new level. They just go into the whole new level of generating image. 29. Download Right files to use in right path for Flux Diffusion model and checkpoint: Download, you just have to go to Flux Dap. You can find Flux Daf fusion model weights here in the folder and you have to save in the CUI model diffusion model folder, and it has a file of 23.8 GB and you just download it. Just download it in flux Snail diffusion model wights. Download this file in CopuI Models Unit folder. That is, go to CfUI open ComfUI Find models. Here is the models. Check for Unit, and just save it here. As you can see, I have already downloaded it. It has the weight of 23.8 GV. Let's download it. Copy this path and paste here. So let's download Flux Snail. Meanwhile, it's downloading. Let's check for another Dave folder. Now, let's check for the regular version. It is a regular full version. So for flux Dave, you have to put it into the CUI, into the CUI models and diffusion models. You have to download it in here. So with the diffusion model, you just click it, open it, and download, check for files version, and scroll down FAX Double. Flex one Dev Sefner downloaded it. In this folder. Just copy the path and page here and just enter download in here. Let's wait for the download. So now we have a smaller version. You have to download it because we are going to use them in upcoming videos. So flux nail we have downloaded, flux Da we have downloaded, regular full version. You can then load or drag the following image in Comfy to get the workflow. So this image is itself a workflow. You just have to save it, save Images diffusion models or save it in workflow. Flux Nail example. And it is a simple use of FP eight checkpoint version. That is, flux Dave has a load checkpoint version. So right click Save Image as Checkpoint example, save it into the CI workflow folder, save it. Flux Snail, right click, save as image, save Now, let's download it's simpler version of flux Dave. It has a 17.2 GV of file and it has a name flux Dave flux Dave let's download it. You have to download this file in ComfUI models Checkpoint directory. That is Go to CUI models, check for checkpoints and put it here. Download, paste it here in the checkpoint folder. Now, let's download flux SNL version. SNL means simply fast. Just click that Download and download it in same directly of Checkpoint. FP eight, flux one Snell FP eight. So let's wait for the downloads, and it will take some time for you to download. You must have a fast Internet connection or I recommend you to download it one by one. Don't download it like I have done it. So let's wait for the download. 30. Work flow for Flux Schnell and Flux Dev Checkpoints: The last episode, we have a sampler. First, we have to add K sampler, add K sampler node, add clip Node two Clip Node. One is negative and one is positive. But adding before positive, you had to add a flux guidance. A flux guidance. For this latent image, I have added a pretty latent image and bought the prompt has been added with a clip. Now we have a load checkpoint. These both are the flux that we have downloaded. These both are flux we have downloaded in the checkpoint folder. After K sampler, add a VAE decode, Latent image with the samples connected, VAE with the load checkpoint. Portrait of a girl wearing hat Harry Potter. Let's Q prompted. Prompt output fail. We have a valid checkpoint. You should have a valid checkpoint that is day one. Let's be prompt it again. For practice purpose, I will share this workflow in the resource file. It might get some time at first. We are not getting any result. So let's cancel it out. Why we will get to know in some time. So understand what we missed out here in this workflow. Check for the images that we have downloaded, and these are the images with the workflow. Let's drag it in here. After dragging it, you can see this image simply converted into a workflow. You can see, drag it and it's converted. Drag it converted. And I'm using flux Snell example, add it here. And we have a perfect workflow here. But before getting into the diffusion model, let's check first for checkpoint. This is the checkpoint, and let's see what we have missed out in our workflow that we have created. We have used K sampler. We have used flux guidance. We used clip and coder. Yes, clip and code negative. Clip code negative ampit SD three latent image. This one is a different node that they have used here. And now everything is same. Only the ampiti AD three latent image is different here. Now let's see what the note says here. Note that flux Tav and Snell do not have any negative prompt. So SFG should be set to one. Setting CFG to one means the negative prompt is ignored. So we have to note that flux tave and SNll do not have any negative prompt. They don't use negative prompt. It. They just simply add a negative prompt here just to keep working with a sampler and they keep the CFG value to one so that it will get ignored. The negative prompt that we have used here, the negative node has been ignored. So let's add. Cute anime massive floppy. Now, let's add. Let's tiger, cute three cartoon style with movie background. Hollywood. Let's Q prompted. It looks nice and you can see the clarity is amazing. And if you add a negative prompt, you will get a image that has no result. One more thing that we have added extra is ampit SD three latent image. This one is different. And within 1 minute we got this image. As you can see, it took 52 seconds, 4 seconds, 2 seconds, 0.7 seconds. These all time taken by these notes, go back to our images and just check for Flux Snell checkpoint. Right now we have use flux Dave checkpoint and flux tape flux SNL checkpoint. Now let's see. This one is also same. It has no change here. And with this, let's read. Note that flux tape and Snell do not have negative prompt. It is the same. But this line is extra here. The Snail model is a distilled model that can generate a good image with only four step. So we have to check for the steps that we have taken here, and this says that the Snail model is a distilled model that can generate a good image with only four step. Let's check it out. Tiger cute, three D with pack ground hollywood. Let's see, and let's prompted. Yes, let's look good. Let's increase the step here, like we have done before. Let's take it to 15 steps. Now prompt it again. The reason SNL is fast because we have decreased the step to just four, and you get the image, and you just get the image. And with just four step, you get this image. After increasing a step, there might some difference in the images, but with just four step, the SNL give you the image. That is why it's fast. We are getting into flux SNL example with the workflow of using a diffusion model. 31. Create Advance workflow for Flux 100+ Styles Image Generation: Now I am loading the flux Nail version because it is faster. Now add a node and search for style. Search for style and search for prompt multiple style selector. Just select it, add this node here. After this, next, change this right click and convert widget to input and convert text to input and decrease the size of it. Yes. Now, increase the space between the nodes after adding the space, now add another node that is just double click here. Search for easy space positive, and you can see this one easy use prompt, select it. Now, after adding this, double click again and select for text. Concatenate, let it. Place here, now add positive to text A, and positive string to text B, and now add this string to text input. Why we have done this, we simply add this tag to a visit so that itself don't take any prompt. It's take prompt from another node and simply push it forward into the K sampler. We are adding extra condition into prompt. This is the prompt that we will add in this, and it will only take just a prompt. And this is the style that he this prompt work according to the style that we will select in this node. And after that, they will just combine into this text concat, and after concatenate, they just give this prompt to this clip text encode and this will encode all the prompt that we have given and push it into the K sampler. And Ksampler will do its work. Now let's add tiger three D Hollywood Q tiger three D Hollywood. Now I have added a prompt here, cute tiger TD Hollywood and add a style. We have the file that is file here that I have downloaded from Internet. You don't have to go through all the stuff and search for it and download it. You simply have to put this into the Style folder in Cofi. Also put it into the main folder in cofuI and check for the main dot pi or nods dot so you can pinpoint this folder. You simply have to copy this here also. Like you can see, I just opened the CPUI and just paste it here. That's styles. And in the styles folder, I have additive. Now, before working into the main workflow, let's save it. Simply export this workflow with stils and confirm and you can see here is the export, Jason, and you can just cut it and save it into here workflow. And now just paste it and you will get the workflow. To work with the style, that's set up our workflow so that we can work with the prompt later on. 32. WAS 100+ Styles Workflow setup: Now go to CPI Manager and go to custom Notes Manager, search for WAS. And as you can see, it has reviews of 1248 try update, restart required. And if you have not installed it, you just have to install it WAS Node suit. Easy use. Install it. Yolan. You can see he is the author and I'm just trying to update it. Try update. You can just install it. Got it and just restart it. Done. Wait for the restart and the new window will pop up after you have copied it, the style, right click copies path. Now in the Cofifolder, look for custom notes, custom notes. Double click search for Was I have too many things that I have installed here. You don't have to go through it. You just have to search for As, search for W as node suit, and just double click it. Now, if you have run the Cf ui after installing the node, you should have this information, these files, and if not, you have to run confuI. Now open the Cf suit file, open with notepad and just Check for the WebUI styles. Now, here the style that is null. And you see here null. You just have to paste the part that you have copied, paste it and double the backslash here. It is important. Just double it. As you can see, there is double backslash here also. Just double it here. Now, just save it. Just save the file and go back to Cf UI. Now, right click. You just need to reboot CfUIRbooto. In my case, I don't know why it's just not getting reboot. So just let open ComfUi and run again. Now, let's check. Here is the style that we have given in the WS style sheet now has been updated, as you can see. If you correctly install it, it will appear, and if not, you have to check if you have missed some step or any step in between. Now with the load point load checkpoint, flux Nell, our same positive prom that have added. And now let's add a anime style here and just add simply Q prompted. That's all. Now see what we have seen here, value, not in the least style. It's asking me for the value. No style, select it no style, select it no style. I have added the value. Now prompt it again. And I hope we get the anime style. Now we have a anime style three D character. Yes, it is kind of three D. Let's remove three D. Now prompt it again? Yes, it looks good. Now, add another style. Painting, acrylic, fancy, illustration, cartoon, Jamat, fashion, sticker. Let's try sticker cute and cube prompt it again. Yes, it's working. Perfectly fine. Now we can create stickers also. It look amazing. Now try to add cat and Q prompted. Perfectly amazing. Yes, it's working. Now let's add. Now you can save this workflow export with styles confirm. I will share this with you in the resource files. It's working perfect. 33. Remove Background from Stickers: New year to you and your family from my side, and I'm really excited to share amazing amazing amazing things with you in these upcoming lectures. We are starting again from the making of stickers, but there is a small catch in making stickers. If you want to create a high quality stickers, if you want to share your stickers, how you can share. You can share them as a JPACO we can say JPG file. Image format that we use all over the Internet or we can use a PNG format, which is a transparent background format. Okay, so how we can remove the background of any image. How can we remove that? For that reason, we have to create another node system so that we can create automatic sticker making machine, a complete sticker making machine I'm talking about. In the previous lectures, I have shared the information about the workflow. As you can see, these are all the workflow that we have used and Snail version, flux day son, flux NL version. And I know that you already have used this version that I have shared with you and you can see a SNL version, flu da version, Checkpoint, and without Checkpoint, how we are going to use and how we are going to create images with them. I'm right now using a Checkpoint version, so that for you, if you have a low VRM or a lower graphic card, you can still use them and work simultaneous with me. And if you have a high quality graphic card just like me, I have a 16 GB of VRAM in my computer, I can use this bottle also in which I'm using a diffusion model, dual clip loader and load VAE. We will talk about that later, but right now we are going to create a sticker PNG format. I have created this version for you so you don't have to work all the way from the start, you just have to drag this workflow in the CFI interface, as you can see. So we have all the workflow that I have created for you, so you don't have to worry about that. Flux SNL checkpoint with styles. This is the style version, like we have already discussed about the style version of FluxNL checkpoint version. These are all the information that we have shaped. And let's check about the cute girl, baby. And let's check for the style. And search for the cute sticker cute. And check for the sticker, watercolor stickers, Illustration stickers and simply Q prompt it. And we have a small error in here. What we have missed out, we have to select the style that is no style. And again, just Q prompt. Now our checkpoint is loading again. Let's wait for some time. So we have a cute baby. It's just a cute baby has been generated. So we have a cute sticker, but the problem is that, as you can see, it just saved the image in your CUI output folder. As you can see, I will show you. This is the CUI, CI window and output folder. It just saved it here. The problem is that there is no background or transparent background. Okay, so it is not just a PNG format. It is a PNG format, but without we need a without background, okay? So how can we generate it and how we can create a copy of it and just automatically create an automatic machine for that. So how we can do that? Let's not waste any more time and you can see high quality sticker workflow, sticker with PNG and upscale, and sticker making with upscale include flux sticker and PNG convert to for business. Okay. So this is the one that I have generated for you and you just have to drag it out. And I will show you what I have done here. Everything is the same. Everything is same. Okay, cute baby girl. I will type it Illustrator. I sticker, cute and sticker Illustration, no style, no style. Got it. And now there is a small window that I have added here. These are two notes that I have added here. How you can add them, if you want to add them, you can do so. You just have to go to manager. In the CFUI you just have to go to Manager, go to Manager, search for custom Node Manager and search for remove PG. And as you can see, there are so many nodes that have been used here. But the one we are going to use is that we are going to use Image remove BG. Easy use image BG. We are going to add this one. You just have to add this into your Cofi After dragging this workflow into your CI interface, if there is any red node here, you just have to go to manager. And install missing custom nodes. You just have to do this thing. Install missing node, and it will show you the missing node and you have to install it. Okay. So there will no problem in using this model. I have added extra node that is remove BG image, and I will save it here separately. So we have two images we will save here and here. One is with background and one is without background. Both are PNG version. Let's save it. And I'm using RM BG 1.4. Why? Because I was actually getting better result using this. So let's prompt it, and I will show you how you can actually generate an automatic machine. Here is a cue, baby. A baby girl, a baby girl has been saved, a perfect sticker you can use, and here is a perfect sticker without background. And I will show you that will blow your mind. A sticker without background. You can print it out and just take it anywhere you want. You can sell it out online. That's all. You just have used some electricity using your graphic card, and you have just created Amazing stickers. Wow. Now you are the graphic designer. AI graphic designer, I 34. Generate 30 stickers in 1 click: At a batch of three and Q prompt ten. It means it will generate three images in one go like this is, and it will run all this process ten times. This is the meaning of Q, and this is the meaning of Badge. The images in one into ten images into three. That is 30 images. You will going to create 30 images. Got it. So right now, I'm just creating only one Q prompt and I'm using a batch of size of three. You can increase the badge size if you want. Right now, I'm just using the batch size of this and just que prompt it. You have to wait a little longer. Why? Because it is creating actually three images in one go. You have to wait for it. And what this node is doing is just simply removing the background of image. That's all it's doing. We will use it separately, and let's see how we can Oh, wow. It's nice. Yes. We will talk about it later. So here are three images, as you can see, one, two, three, two, and three. These three images has been generated, and you can see across here. Just click it and you can see all three images. You just click it, click it, click it. And actually, I love I really, really love these stickers, actually, and you can sell it online. 35. Create high quality stickers or images about 6000px: You can create 1,000, 10,000 lags of stickers according to different characters and according to different genre. But if we want five MB or six MB or the size of the images, like in the red bubble, if you are going to upload a sticker, you have to create at least 5,000 X 5,000 pixel of image or PNG. So they will accept it and they considered it high quality PNG. How we can do that. For that reason, we are going to use another workflow that I have already created for you. That is sticker making with upscale include. Let's drag it out here. Now here is a small catch again. Everything is same. Lot checkpoint, IostPanda Stiker with pink background, same Illustration, children book, same, everything is same, everything is same. Everything is same. What is the difference? Here is the difference, you can see. It is the ultimate SD upscale that I have added. And here is a load model. We are going to use this four X NMKD model, and we are going to use clip encoder. You don't have to write anything here. And after upscale it, we are going to use our node again to save our image without background. So let's start again. First, this workflow is same as you can see, this is the same workflow that we are using again and again. But in this version, we are going to save this as a image about 1080, 1080 or 51500 X, 1,500 square images we were using. But right now we are using a upscale version here so that we can upscale it by three. As you can see, here is a upscale by three, and if you are going to multiply this three, with this 1024, 1024, let's check for the result. Let's make it 1,515 hundred. So I just created a Latin image of 1,500, and if we multiply this by this upscale by three, we are getting the So for the RAD bubble, if you are going to upload it on the RD bubble, we are going to use we are going to change the dimension of our Latent image is 2000. And again, 2000, we are going to use this 2000 2000 Latent image. And with the ultimate SD upscale, we are going to upscale by three. So 2000 into three is 6,000. It is a simple calculation. In the Internet, if you are going to use 6,000 X 6,000 pixels, it is actually a really high quality image. You can see and you can use that image anywhere in the Internet. You can upload that. You can sell it. You can create smaller version of it and sell them according to size. You can create high quality stickers and sell them on Red bubble or As website. Let me generate a small, small, small image for you. Let's cute inositPanta sticker with pink background. Cute sticker, no style, no style, no style. And right now, just que prompt it. You just que prompt it. And let me tell you one thing. It's going to take a lot of time. If it is generating a high quality image, it will take time. So at first, it will show us the image that what it has generated. Here is a panda. You have to wait a little longer for this because upscaling the image to this extent will take a lot of time. So generating the image is not that bulky. Here's a sticker that has been generated. The second part of our workflow is in the action. How it's working, it's simply increasing the size of your image tiles by tiles, as you can see, tiles. These are the width of the tile and height of the tile. These are some complex terms that we don't have to understand. We just have to wait how it's working. That's all. As you can see, it's working tile by tile, and it depend on the complexity of the images also. I'm just fast forwarding the process. So here is the high quality, you can see. What is the catch here? Let's see refresh it. Now, as you can see, it has a 5.83 MB of size. And if you are going to zoom it out. So we have created high quality images without spending Rs, sketching or editing. And you can see it has a 500 KV and it has a 5.8 MB of size. It has actually a big size. And I hope you love this lecture. See you in the next lecture. Thank you for watching. 36. Understanding ControlNet and different preprocessor: How it works. As you can see, there's a stable diffusion model and here is a control net. And as you can see, there's a time decoder, text decoder, encoder, time encoder input, encoder, encoder, encoder encoder. There is a output. So we just input the image prompt, and it will process middle box like a K sampler, as we have already discussed in our previous lectures, and about this, there is a output. So all these are information that is encoded and then decode and then we have output. So now here comes a control net. When you give the input, it just process, go to K sampler and with this middle works, a control net, A condition transform, is a condition that has been generated. It just encode all the conditions and after that, it will give its own output which will after zero convolution, there will be a decode and we cut the image. This is how stable diffusion and conton at work. We are not going in a technical aspect because we just want the result. Leave all these technical things to the engineers who are working behind it. So what are the advances about the model? Use bucket training like novel AI can generate high resolution images of any expected or use large amount of high quality data. The data covers diversity of situation. Use recaps and prompt like Dali and generated Good. You can read it out and understand it one by one. We design new architects and that can support ten plus types, Tmage to image, text to image. Thank you very much. Hg information. These all are the variation that has been generated validator, tile de tile super resolutions. You can see it just increase the ratio of the image, open pauses open pose, different type of depth, open pause, CE this is known as nE This is canny line art. Anime line art, scribble, PD, soft edge. All these are technical terms, open pause plus any. Control means we are going to control our image generation. It's name itself explained much to us. This is known as Control Nt. Now let's use it practically. See you in the next lecture. Thank you for watching. Thank you for your really, really night reviews. I really love 37. ControlNet Aux Preprocessors Different processors output of images: Use the Control Net node, we need custom notes section, and we are going to install that. First, let go to the manager and go to Custom Node Manager. Now search for the term Venture. Search for SI PH E XYZ. Try update. If you have not installed it, you have to install it. I just updated it, restart and restart it. And after that, search for pre processor. Comfy Control Net auxiliary preprocessor. Try update it. I've already installed it, I just updated it, and if not, you have to install it. That's all. Now, go to Com f roll and search for Cf Studio, Szel TrydtO install it. So after installation, all these nodes, just click Restart. It might take some time. Now in your CI interface, go to dit, clear workflow. Now, the workflow has been cleared. Now let's add a node. Search for CtrolNtPre and you can see Control Net preprocessor. Here is the name, art venture, loader, click it. Now you can see here is the Control preprocessor. So this node has one input of image and one output of image. Just click drag the point out of it and you can see you have to select the load image here. Once you load image, load the image from my CofiFolder automatically and on the right hand side, drag it out. You can use either save image or preview image. I'm using save image here, so our image will get save automatically. So now let's Q prompt it. Tu it just process the image as you can see, we have got really nice amazing result. I'm just kidding there is no result. It is the same image that I have created using the same image that has been loaded here, it just drag this image out of this without any processes. Why? Because there is no preprocessor has been selected. So we have to select the preprocessor. You can see there are many, many, many things has been it. As we have talked about before, there is a scribble in art, animal in art, A PD, MS SD poses. As you can see, all these poses, all these processes has been added. So there are many processes that we have already discussed a little bit of overview that we have talked about in our previous lecture, canny scribble, MLSDPoses, open pose. We are going to use all of them one by one. Now let's select for the canny. We have canny processor overt, none, and just Q prompted. Now you can see here is the Ky selected, Cy prey, Q prompt. There's a difference. Line art, Q prompt, here is another output, scribble prompt, here is scribble. MLSD que prompt, here is MLSDOput, open post, open post will not work here. Some preprocessor will not work depending on the images. You have to select the right Control net preprocessor for that specific image. You have to experiment it. Now you can see here is the Kenny result we have what preprocessor has done with the image. It takes out the information of the image, how image look using the exposure and shadows of the image. As you can see, there is a exposure. And it just create a exposure out of it. The object has been highlighted in this image and the background is separated. So it just check for the edges, the outline and the depth. Depth means shadows, exposure, and the distance of the object and background. How far the background is from the image or from the so it just process all the information and give us the result that we will going to use in our AI image generation in our upcoming lectures. So you have to experiment with the resolution also. Once you increase the resolution and try to reprocess it, it might give you a better result, or in some cases, it might not give you a better result. Check it out, can he? So there is a limit of the resolution you have used for the Kenny. So you have to check which resolution works fine for you. So now we have understood, as you can see, here is the pose. He is the pose, but it's not working. Let's check for another image. No. No. Let's check for the pose here, CP as you can see, here's the pose. So pose is like a stick figure. It just use the pose of the image according to the face, it detects a phase. And with the face, it just check out for the hands and borders of the image, like these are the borders. So solders, it detects solders like a borders. He is the body part, is the hands, and here is the hands. So you have to experiment it. You have to understand it on your own. Select for that dep prompted, and it will give us a depth map for the image. So it might give it a bright area and dark area so that the AI will understand the objects and the brightness of the objects, how far they look. In this case, as you can see, the brightness, the contrast, the background, and the image. It's working really, really well. I hope you understand the basics of this control net processor. In next lecture, we are going to use these detail in our workflow. See you in the next lecture. Thank you for watching and please give it a five star rating. 38. ControlNet and Flux output Comparison: So after the update, the problem I faced in the CUI is that it's not working with the SD Excel or ST 1.5. I don't know why, what is the reason about that. But the case sampler always show some error during the process. So I have started working on flux, and it was working fine. And the results are really good better than SD and ST Excel. So we will talk about flux and how to use and how to use Control Net with flux. So as you can see, on the left hand side, you can see it is the original image that I have used, and on the right hand side, with the flux model, it's generating really nice results, as you can see. I have created more results using this image and some more image I'm sharing with you. You can just check it and understand how these things work. I will share the workflow and how to use that workflow, I will share in upcoming lectures. You just have to sit tight and understand that workflow. Meanwhile, you can check that generated output. All these things were created with different settings, we will discuss all of them in upcoming lecture. 39. ComfyUI Advance workflow for Control Net Manipulate your images: Hi, my friend. Welcome back, and thank you for taking this course. I really appreciate. And if possible, please rate five star to this course. It will help me. Right now, this is kind of a web and it feels scary until I will explain it. What you have to do first, you have to go to your resource file and simply drag this into your CUI workflow like this. And this is what you will see. This is the workflow that I have generated for you. It is a simple flux control so that we can control our image output according to the image we will input. This is Tazml from India, and it comes in the seven wonder of the world. Once you load this workflow, what you will see, you will see two things. First, you will see a bright pink color of node here and somewhere here. Why? Because the node which I have used here, it is not installed in your system, so you have to install it. So how you can install it. First, you have to go to Manager. In your CFUI workflow, go to Manager, click Install missing custom node. When you click it, the nodes will appear here. Right now, there will be no node which will be going to appear here because I have already installed. But right now it just checking if there is node that I have missed out or not. So once the node appear here, you have to install that node and restart your system. If it's not restart, then you have to cut it again like this and run your CUI from the start again here, you have to run it again using NVDA GPU and if you have a CPU, then you have to run from CPU. After installing that, you have to go to install custom nodes. I take a scenario and if you don't get any missing node here, then you have to go to custom node Manager and search for X lab sampler. Search for XLAb. Let's check it out. Custom node. Wait for some time and search for Xab. Here is the X lab node. Mixed X lab nodes we have X flux CPI, you have to install it. You have to mix lab, you have to install it. You have to install both of them, and the node will appear here. And if not, after installing that node, if it's not appear here, then you have to Double click Search for xlab, check for the node, which I have used here, and you can see this is the one that I have used. You can see and just click it. Here is the node. By the way, you don't have to go that far. So this workflow will work perfectly. Don't worry. I have already tested it. And after that, whenever you run the prompt, what will happen your prompt will stuck here and here, what we have to do. In that case, as you can see, here is a flux one DEV quiveks dot GGF. You have to search for it. Example, Flux one day four, our quarter or quarter. Anything you can use. This has a value of till o qu eight. If you have a higher VRAM or higher graphic cap card, then you can use qu eight, which might take some time. So let's search for it. You can see quarter, odqter, third quter, our qu eight, just click it. This all our value has been appear here and you can download it directly. So how you can download it if you go to source file, Hugging past dot Man and search for GGUF, you can check quarter, odqterthquter, QqourQqar o qu aright, and right now we are using quarter five. As they have already mentioned here, too, where you have to download it. Once you click Download, you have to go to ComfUI Models Unit. Let's check it. Go to Cou Models. Search for UT. Here's a Unit. As you can see, I have already downloaded it, and now once you run this XLlab sampler one time and it will give you an error, what it will do, it will generate a folder in the name of XLAb. But first, you have to run it once. Even if it gives you an error, you just have to run it. Once you run it, it will automatically generate this folder in the CPI model and just click it and you can see it has a Controt folder. You just have to double click and you can see there is a file. Flux dap Control net V three saf ener, and here is the file. Load flux Control net. Here is a control flux dap control. As we have already discussed about the auxiliary process, what it will do. So to process it, we need a flux depth control net, which specifically work with the flux. And let's check it and search for depth and run it Let's see what we get. Here is the control net debt available. And in the same image, what we will get, wait for some time, and we got nothing, as you can see. Here is nothing we can see. Why it's happened here, we have to play with the settings of Control Net. Let's make it seven. Okay. Make it denise 0.8 and run again. So how you download this flux deep control net, you have to go to flux Depp Control Net, you have to go to flux Dp Control Net V three, search for it. You can see Hugging face has the first link with the XLlab. Here is the demo that they have shown. Here is the file and download this. Where you have to download, I've already mentioned. You have to download it in the X flux in the Control net. See, we are getting result. Now, let's go to 0.4 strength of the image and 0.5 for the denoise, and let's check it out if it will work or not. Wow. Now, this is the image generation that we got. But for this, we need to go to five and let's check it if it will work according to the image that we want. And I have increased the value here, increase the strength of the image to control it. As you can see, we are getting somewhere, as you can see, in the forest, building covered it with grass, and here is the strength that it's using. Let's increase the strength further 0.7. And Q prompted. Yes, we are getting somewhere, and right now, 0.8, one, six, and Q prompted. You get the idea, right? This is the basic workflow for the control net using flux. 40. Advanced ComfyUI Techniques High Quality Outputs with more control on Depth Map Control Net: And once you go to XAV Control N taped, here's another workflow, you can see. Here's another workflow, just download it. And once you download it, I just download it in my workflow folder, Control Tab volume three and just go to your interface, press this press Plus button, click, go to Control Net Tab workflow and paste it here. And here is the already prepared workflow you can use. And if you get any of the node in the bright pink color, you just have to go to Manager and install missing node. You just have to click here and just upload an image. Let's input this image open. Here is the image, and now just Q prompt it. It's creating a dap map of the image. Midas tap map. Now during this process, I got this type of error. Here is the error that I got. So here is the diffusion model. So we have to select for the diffusion model here. Let's select plug Tav. Here is the error. Let's prompt it again. Plus Dave Sapner It's loading the model. It's now working. I just change the diffusion model here. That's all. That is the flux. And right now, XAb is working its part. And now XLb is working. Wait for some time. This was the image that we have used. And this is This is the image that we got. And I loved it. I actually loved it. The way it detailed this image is really amazing, really, really amazing. As you can see, Wow. Now, let's BEA, beautiful. Woman Wait, beautiful woman in fast red dress style fashion work image. Now I have used our prom beautiful woman in red dress style fashion. Got it. Now, let's prompt it again with this prompt. I save this workflow for you. Export workflow. For course. You can find it in the resource file. I will save this workflow and share with you in the resource file so you don't have to be worried about the workflow. We are getting result. I don't know if the hands are okay or not. This is the workflow. Can you understand? I start from this sampler. Like K sampler, there is a X lab sampler from third party, and you can see we have connected model. We have connected conditions. Here is a clip, and we have connected latent image like we already did in our basic model. And with that, we connected control net condition. He is a control net condition. With this condition, there is a control net which control the depth of the image. Here is the image, and here is the image that has been generated using extra Dapmp. Here is the image image. This is the image goes here. Here is the control net of the image, and here is the model which use the control net to create this tab map here. And for using a diffusion model, we need to use a dual clip coder here is the extra clip encoder, which has a prompt, a negative prompt, a positive prompt, and here is the image preview for the DAP map which has been created, and here is a preview image, here is the saved image of the tap map. And here is the diffusion model. Dual clip encoder, diffusion model. And here is amply latent image. If you check thoroughly all the workflow, you will see all these notes are using in a common way. Results are good, but hands are not. Let's again Q prompt it and wait for some time. I simply jump into the future. Once Q prompted, the workflow is not working. Why? Because we have a noise or seed, which is fixed. So I just make it randomized and now Q prompted, wait for some time, and now I just jump to the future again. We got the result. And now results are good but hands are not good. Love the result. You can see the image. Now let's check another image. I just check it out here. Now cue prompt it again, that's beautiful woman in plu, magical background. Just click it. And I think there's a cue prompt, and I don't know if it's work or not. And here's a beautiful woman in flash magical background. And if it will work or not, let's wait for it. Wait for the result. I think it's not getting the result. And I think because of the prompt that I have used. So let's prompt it again. Now I have used beautiful woman in blue fashion magical background. And let's see if we get the result or not. It's tried to create or recreate it. And it has a third hand. Spooky one. But yeah, it's try to create that, as you can see, it's tried to create that, but I think it's a bit difficult for it to generate. So let's again, try to use a portrait. Let's try to portrait of this image. So I'm just taking a portrait again, a well defined portrait of our image, and now click the prompt. And I hope this time we get something. You can play around with the image dap map if you want. The dual clip coder, you have to select for the flux. This is the important part. Once you understand the settings, how these value work, you will get to know how you can control your output every time. I feel like I have missed something. Why? I feel like I just have to sitting on the table, sitting on the table, holding pen, purple, right now the result is good, but not that much. Now let's check with another prompt. I'm taking so many examples so that you can understand properly. And you will get to know the idea how to play around with the values, how to play around with a prompt, how to play around with the image. That is why I'm using so many examples to explain it to you. Right now, it is working with the tap map only, and in the upcoming lecture, we will going to use different type of that map, Kenny edges, and many more different type of control we are going to use in our upcoming lecture. Understand them one by one. Can you see hands are same, almost the same. Results are somewhat yes, same, but I really don't like the result much. So yes, results are somewhat same, yes, as you can see, like the dap map is generate a really good image, but it's not eye catch image. Let's try to regenerate, and I just simply change the forest magical to magical forest background. And I hope this time we like it. With this result, hope you get the idea how this prompt work and how this workflow is working. And now let's dive into another part of this control net module. 41. Overview of creating our own LORA: Before moving forward, let's recall that this is what we have created in our last lecture. And after that, I just created, Oh, this is me. I just come from the space, and right now I'm just wearing the astronaut and I don't want to share my personal project here, but it just come out in front of you, so I have to share this. So in the upcoming module, you will learn how we can create our own Laura and use that for personal and for professional use, and you can create your own Laura using some different type of image patch. This is me, and you will learn in the upcoming module how we're going to use this Laura. I just experimented it on myself can you see how handsome I am? This is me in the war. This is me. Wow. I'm just flying. And this is my side business. This is my side business, and I'm also superman. This is also my side business, and I'm Iron man right now. This is me in the Tose world, and behind me, it's Nebula. This is me, handsome me. This is me. Can you see how handsome I am. And I have six pack. Yes. This is me, my muscles, and my under arms. This is also me, me, me, me, me. This is me, handsome me, beautiful me. This is me. Son, this is me. This is me. This is me. This is me. This is me. This is me. This is me. This is me, me, me, me. And it's look actually real. The body, can you see the body? How it just merge everything. This is me. This is not me, not me. And this is not me. This is beautiful me. Can you see how beautiful I am? This is me. This is I just created anything. Can you This is me me me me me me, and can you see? Yes, you can use your image, your own flux, your own Laura. And use them in any kind of thing. So in upcoming module, we will learn how we can create our own a and how we're going to use that Laura with flux and create our images. So you can play around with these images yourself. We will talk about that in upcoming lecture. This is just an overview. In x lecture, we will learn about CNET I will share a workflow with you, which will be very handy. And in upcoming lecture, we will learn how to use all these Las. Okay? So in Lx lecture. Thank you for watching. 42. Advance ControlNet workflow all in one Depth map: So this is the workflow that I will share with you and you can download it from your resource sector. You just have to drag it like this and it will appear here. There are a bunch of nodes that you have to install. Don't worry, we will learn that. This is the workflow and don't scare out of it. Okay. This is really simple to use. So first, you have to understand how these things work. It just load the image. It just process it, process it, process it, process it, process it, and then now come clip text code where we put our magical prompt. After that, it just c the control net how the image will look like. It just get the overview or the sketch of the image. After that, it just created the image here. And after that, it will just upscale the image for higher quality. So let's don't use it right now. This is the unable and disable part. You just have to click Unable or disable. You can do it with anything. Okay? So this workflow is really simple to use. There are going to be so much nodes that it will install. So once you click Manager, Install Missing note, there are going to be so many nodes you will go to see and you just have to click here and install one by one or install here. Install it and restart your comp UI. That's all. After that, so we are going to load GGUFFlux Model. Where you will going to get that to install it, you just have to go to Manager, you just have to go to Model Manager. You just have to search for qu five, and you can see you just have to install one of them. Don't use flux one Snail version. You have to use flux GGF. You just have to match this name. As you can see, this is the flux D quarks dot GGF. You just have to go to Manager, match this name, install go to modern Manager, flux q quarter. This is the name that matches correctly and you just have to install it. That's all you have to do. It might take time depending on the size of the file. Right now it is eight GB file. You just have to install it. Simple and very easy to use. Okay? This has a range of quarter to qu eight. Ir qu eight is the best one. If you have a higher VRAM, you can use o qu eight. And if you have a lower VRAM, I have 16 gigabyte of VRAM, so I'm using a quive one. Okay. And if you have a lower version of VM, you can use lower version of GGUF model. After that, you will go to see a red mark hair, a red stroke or a red border hair. You just have to copy this name again. Search for flux slashaedt SafNer Manager, go to Model Manager and search for Flux one. You can see, this is the one that I have installed, Flux one VAE model 335 MB, VEE model, ae Sefner VAE. And let me tell you one thing. Please don't just install anything from Cf UI. There are some hack is going on in Cf UI, right now. So you just have to save yourself from Cf UI. Hacking software is going to just malvey PC. You just have to be sure you just copy the same exact file name that I'm showing you. Those are safe to use, okay? Okay, so you just have to install that. So get back again. And this goes same for this file also. You just have to go to Manager. Model manager search name or just copy. Just check the name and just install it. It's very easy to use. And after that, there is Laura, Laura, and there is prompt. Now, same goes for this also clip L. And if you want to enable the Laura here, you can do so. You just have to select this Control plus M. That's all. You can enable or disable it or you can just right click Bypass, right click, Bypass. That's all you can do. Another thing, you just click it and Control B, you can see, easy one. And it is just an optional, you can see here. It is just an optal you can load it. And if you don't want it, you can load it. Okay. And make sure flux is selected here. Type is flux. You have to select. Now, as you can see, there is a lot more thing you are going to see here. And once you check it and uncheck it, you just have to select one. You just have to select this one. Can you see flags on Saka lab Controller Unit Pro diffusion, Picho M you just have to download one Controlt Load Control net model. There are so many control we going to see here. And you can find the flux SECA lab controle Unit Pro diffusion part and you just have to go to manager. Copy this name SECA f Control Net Union Pro. You just have to search for it. Again, Mode Manager. Saka lev Control at Union P. I have installed it already. You have to install it. That's all. As you can see, it is installed already and you can get back again, close it. And if you can't find that, you just have to go to your workflow, and you just have to rephrase it. That's all. And if you don't know how to repress it, refresh notes definition. That's all you have to go. A, refresh. And now we have already set up everything. You can check this out, check that out. We just set up everything. And now the prompt model girl pretty with magical background. No BEAUTFllGirl, in the magical world of butterfly. Now, let's Q prompt. Without ya, we are using. And right now, I just bypass and I have bypassed the upscale factor, so you don't have to wait long. And if you want to upscale it, you can upscale it again. Let's wait for some time. The loader is loading. Let me tell you one thing. How are we going to achieve the image that we have used? How are we going to achieve a similar effect is by using this strength, start and end percentage. As you can see, the strength is I'm using right now 0.7. Percentage of start is zero and end percentage is five. Have to experiment all three values with your image. Once you get the hang out of it, once you understand the basics of the stand, start and end percentage, and you will learn how to use. Once you play around with this value, you will get to know yourself. And this is the thing that no one can explain you easily. You just have to experiment it and you have to understand it. Once you create about ten or 15 images using this value with different different values, you will get to know how all these things work, and you will get to know how and what it will do for you. Okay? Let's wait for some time. You will going to love this image. So as you can see, this is the result that I got. Using this image, this is the result that I got. Can you see? It's really amazing. And now after that, once you use this type of Control N, you can just disable it and unable second number of control net, and you can play around again, with the value. And right now, I'm just using and percentage 0.5 strength 0.9. Let's Q prompt it again. You just get a line out of my image and wait for the image. What will happen next? This is the image. This is the image, and this is the image. Let's compare them. And this is all the output that I have used. This is the image. This is the image. Can you see the pose is same. And now after that, I've used another image. Output. This is the image. This is the image. This is the image, I just created using this image reference. This is the image and here is the reference. Here is the reference. Here is the reference. Here is the image, and this is the image it just created. Recreated, this is image. Recreated image, here is the image. This one is a recreated image. This one is the image. Tis, Tiss. This one. This is the image. It just re created using this image. You can see this is the Tasml that it just recreated the Tasmlimage we have used before, recreated, recreated, recreated, recreated, recreated. This one is the image that I have used. This one is the image that I have used. This one, this one, this one, this one, this one, this one, this one, this one, this one, this one, this one. We have discussed all these things before also. So right now, let's check for our image, and now it just created another image using this structure. You can experiment with all these things. You just have to uncheck, check and just use La. And let's check for the Laura. Let's bypass it. Now let's prompt it again and wait for our result again. Wait for it. We will get in some time, let's jump to the future again. So here is the result. It's look somewhat similar, but not that much. But let's understand the basic structure and create a image out of it according to the liner that it had created that we call as Kenny Control N. 43. Download right file to create you own Lora for Flux: Create your own flux, first, you have to go to Pinocchio. I will share the link in the resource panel. Jim, or you can just Google it. Pinocchio flux Jim. That's all. After that, you have to go to Download Pinochio first. Once you click it, a new window will pop up and you only have to click the Download button here. After that, it will take loading and scroll down Download four Windows. It will download, and it is around 100 AmboFle, so it's downloaded I just open it, open it, and run it. Simply double colligate and Window will ask you to protect your PC. You just have to go to more info. After that, more info. Run anyway. It will take some time. Now a window will pop up and you only have to allow Pinocchio to make the changes in your system. Windows will pop up. You can select your theme light or dark, Save as a window will pop up. After that visit discover page, you can check all the information here and start downloading as per your need. But right now, we only need Flux Jim. You can search it on flux Jim, FLUXGYJim. This is the program that we needed. Click it and start downloading it. After that, in the new window, scroll down, install. It will take some time, not sometime. It will take a lot of time. EQT is downloading a bigger files, like we have downloaded before. So it is around 2021 GB of file. You have to wait. All the process you can see here, as you can see, it's downloading and once it download, a new window will pop up and you just have to save as fluxjym dot. It is just asking you where to download your files. That's all. You can save as Get and download it. Again, a new window will pop up, flux Jim and start installing. Now it's again downloading many files for you, wait for some 44. Steps to Create the lora how to run right way: So after downloading, you will see this screen. So first, you have to just screen this size, and after that, if you want to terminate it, you can just stop it here. Let's open it in web, open WebUI pop out, clone, pop out. When you click Wop O, it will just open in your web browser like UI. So it has its own user phase. So in Jim and publish, here is the Laura and here is the Laura Info. What is the name of the Laura? Let's name it Karen. Transcript. Okay, just working. It's transcripting work, so you don't have to worry about your Laura. Triggered word. Word or sentence. Let's make it current. Now, base model, I'm using flux, Snell or Dev. It's not working well with Snell, so we are using flux Dav. So let's check for. So for the VRAM, I'm just using a 16 gigabyte of VRAM, so I'm just checking it. I have a 16 gigabyte or if you have a 12 GBO gigabyte, you can use this. Repeat trains per image. Make it simple, simple, don't check it. You can just experiment with these things, expect training steps. You can Nothing. Just leave it blank. Simple image prompt, separate with new lines, current. Now sample image every new N step, re size data size image. Let's keep it five, one, two, Y because of the size of the image. Now let's upload your images. Simply, you just have to select your images. Now you have selected your images, you just have to check for step three and start training. That's all. If you want, you can take another step. But before you start training, you just have to click on the ad AI caption with Florence too. You just have to click it, install it, and it will generate a caption for all the images that you have used. After that, you just have to click Start training. It will take some time, not sometime, it will take around three to 4 hours depending on the power of your machine and depending on the images, how many images that you have used. It will take a lot of time, but the time you invest, it will worth it. Now, if you want to check the safe file or the Laura that has been created by the training use Influx Gym, you just have to go to publish. And in the published section, you will see a link. Trained La link. You just have to copy this path and go to that part and here is the output folder. Once the training is done, you will see many files will going to appear here. As you can see in the dataset, flux Gym data set. In the current section, you can see all the training has been generating and the file AMB ten MB, 15 MB, you just have to wait for the training to be complete once the training is complete. You just have to click the sample. In my case, sample was not working. If it's working in your case, it's good. But right now it's not working in my case. So the training has been completed. Now let's check our La that we have created. So in my C drive, I have created the folder Pinocchio. I double click it open. Now check for API, flux git, double click the output. This is the folder in the name of current has been generated according to the oa. Now, double click it. And in the output folder in the current folder, you can see current safe third. So this is the loa that has been created. I just simply copy this loa that I have created and paste it into in the download folder in the CUI Window portable, in the CUI, go to models, select for Laura and simply paste it here. As you can see, I have already pasted the Laura in here and just run the Cf UI and use and use all the workflow that I have used before. You already have the workflows. Now you can use your own. 45. How to use Lora you have created: So this is the workflow. You will find this workflow in the download link that I provided you before, and you can also find it in the resource section. So you don't have to worry about this workflow. And I feel like this workflow is really amazing. So first, you have to check for the Laura, as you can see here in the CR Laura Itech. You have to select the Laura here, the current Laura that we have generated. And about this current Laura, you can download this Laura and create images of myself, and you can share it with me, and I will only laugh after you create this image of mine. And I can't wait to see all that image that you have created using my face Jokes apart. Now, you just have to normally like you select Laura, you have to select this Laura again. Okay. So run the prompt here after prompting, you just have to click Q prompt, that's sold. And all the images you will see here, it will just run let's check it out. What it will give me. Yeah, I am become astronaut. And after that, I just paste the prompt. I just copy the prompt from the hat GPT and ask JA GPT to create something amazing for me in the forest. This is the prompt. You can just copy this prompt. And here is me standing alone. After that, I have tried so many times and run this confui on my face. So here, as you can see, here are the images that have been generated. So many images that I have created, Iron Man, Superman, so many. And this is me, my body. This is in the Buddha phase. Actually, I have tried to create something artistic, one, but it's not working well with that. But after some hit end trial, I just finally got my female version and you can see, it's getting better and better. As you trial. This is all about trial and error and the Epanion based thing. But one thing I understand in this creating Laura is that the more images you use to create your own Laura, the different type of imagery, the more it will get better, that's sold. And I hope you understand what I'm trying to say. And I really thankful that you people like this course, and I request you to please give it five star. And your positive review, I really, really appreciate it. And I really try to update this sures on the regular basis and share the latest technology with you in Come PUI. Thank you for watching. See you in the next 46. Install Raster to Vector Nodes and Basic Information: Welcome back to this CUI course series. So today we are going to generate vector images, and this will going to SVG vector files, so you can scale it out as you want further and further and further without losing quality. So first, you have to go to manager. In the CPUI interface, you have to go to manager, custom node manager, and you have to find this node that I'm typing here. Now, search for SVG. And install CPUI to SVG, Yanick 11, two, and install it. Wait for some time, and now restart it. And now you have to confirm you want to reboot, reboot the server, and confirm. Let's wait for some time. Now it's reconnected, get back to it. Close. Now double click on the canvas. Add SVG. And as you can see, there is a SVC preview SVG preview node. This note allows you to a preview of your SVG file before saving it or using it. It helps you check if everything looks correct. Save SVC node. This note lets you save your work as an SVG file. SV files are great because they don't lose quality when resized making them perfect for logos, vector based design. Rester to vector node. This process convert an image made of pixels into a clean scalable vector. It's useful when you want to turn a blurry image into a sharp, high quality design, vector to raster. This convert a vector image into a rester image. This is helpful when you need to use a vector design in place that only support pixel based images like photos or certain digital artworks. Raster to vector. This is a special type of raster to vector conversion that simplifies the image into only black and white. Make it useful for logos, dances, and linear. Raster to vector node that is in color. Unlike the black and white version, this method keeps the color while converting a raster image into a vector format. It's great for creating colorful logo, high quality vector design. 47. Create Black and White images to SVG files simple workflow: Now we are going to use our register to ctify. Let's select it. Here is the node that selected. And now, Dole click and load image flected add this image here. So these both nodes are connected, and now and now let's add save node. Means SVG and select for save SVG or PV SVG. Selected, done. Save SVG, Image background, SVG string. Now these are connected. Now let's load a image, drag it here and if we just load it, and now let's Q prompt it. Once you que prompt it, it will just run it out as fast as it can depend on the processor. And for me, it just take 0.2 seconds in a blink of a second, it just created my SVG file. Let's check it out. But we have created in the compu folder, find your output folder and check for SVG. In the SVG file, you can see it is SVG file and how you can check if it's SVC or not, you just have to check for illustrator. I have Illustrator. Let's understand how all these things work. Okay? In the Photoshop, as you can see, Photoshop works on rasterization, which is sed image. So Photoshop work on raster images. As you can see, it just raster and pixelated. This pixelated is known as is and into our AI, that is Illustrator, and as you can see, once I increase or scale it out, it will not register. You can see it rested. As you can see, this image is pixelated. This is a simple JPaC image, and this one is scalable vector image or scalable vector graphic. It's pretty cool that we can easily create our images into vector. And if you don't have Illustrator or oral row, you can use CF UI for yourself. We can do this within Illustrator also. Let me show you how there is an option in the Illustrator image dress. Once you image dress, click it and it just created it, as you can see, it created into a vector. It is very simple image. It is not very complex image, so it is easy for it to change it into vector size. As you can see, both results are pretty good. Here is the SV five. This one is that we have generated within the illustrator, and this one is we have created in CFI UI. 48. Restor to vector and Different setting for better results: Now, double click on the canvas, select for SVG, and again, select for vector to raster. Select this with the SVG string, and now again add another node, double click Save Image. This will going to save your image. It will convert back your image into raster image, and your AST image will be saved here as a PNG five. So just run it again. Here is your raster image. And in the CFUI refresh. Has your raster image. Hazo S VGF. Once you select the mode, they are different mode and once you polygon it and run it again, can you see the difference in both of the images? What it has done to your images, you can select polygon when you have a straight line design, like here, they have a straight line design, and it will create nice and straight line for you for the spline, it will create a smooth curve save in the design. There are more settings available in this. Once you work with the complex images or more color images, it will help you in 49. Ask ChatGPT for prompt the right way: Let's load the workflow that we have used before. Here's the workflow for Flux tape, bring it out here. This is the simple version that we have created before. As you have follow along my courses and curriculum, you will see this note. And if you are lazy enough to go back to your workflow and find it there, simply download from your resource file, the link that I have provided you. And for this purpose, I use hat GPT, go to hat GPT, and ask hat GPT, give me prompt for black and white, cute dog prompt for stable diffusion without. And in the code, you snip it for easy copy and paste. Paste it here, run the prompt, wait for some time, and this is not the image that we want it. So let's ask chat hPT. I want it in the alone. Style. And if you have not got the images that you have desired, you just have to tell the chat GPD and It it in the slot design or style, just copy it and paste it again and prompted. For that, this puppy is actually cute. Let's wait for some time again. No, it's not working. Background white and clip art style. Copy it again, paste it here. After that, prompt again. And I hope this time we get it according to Slott of the black shape, we clay part, high contrast, minimistic clean edge, tie style, simple and pool design. Yes, now we got phase. I convert it Q doc phase. But right now we have a QTc but I just addit QTc phase. So let's Q prompt it again. It's give you sometime blurry images, but for the day version, you might see this type of images which looks blurry. You can reduce this by steps, goes to 30. And now, again, Q prompted. These are all the problem that you will going to face, and I am showing you all the problems you might face and how you can rectify all these problems that you will face. Is the pace, but there is no eyes and mouth pace with eyes and mouth. Remove fork, ball design, simple cartoon style. I have added some prompt here, as you can see, there is a cartoon style, vector style simple. With the increase in the steps, it might take extra time in the generation, but prompt it again with the style that we have added now and here's the image, which is not blurry now, wait for some time again. And let's see what we got now we have Got this image. And now let's convert it into vector, and let's add some node in this work flow itself. I will share this workflow with you so you don't have to work again or generate all the workflow, all your bi. 50. How to upscale any image : Shared some inside of the workflow, like how you can only upscale the image. And I believe it is important for you to understand how it works. Check for the resource file. I have already shared it with the resource section. You can check it and download it. You simply have to drag this workflow into your Canva. Now, this is the workflow that I have already created before. What you have to do in this workflow is that you simply have to go to Sampler Custom Advance, click it Control M. Now it's enabled, and once you click Control again, now it's bypass. Or we can say it's disabled right now. What you have to do is that you only have to choose a file here, choose file to upload, select this file, select any of the file, just click it. Once you click it, I'm using a SASke that I have created before. Once you click it, click it open. Now the file has been selected. It is a PNG image that I have created from one of my workflow. We have discussed about it how to create a PNG and how to create a stickers. And I want to tell you we will discuss more about it in upcoming lectures. So what you have to do is that simply remove text I think we need a prompt here and upscale. You can type anything. It won't affect any of the picture. Once you click it, and here you can see the settings. Upscale by two means the size of this image will get double. And if you want to increase the size, you can do so and just click it by two. Now, leave all the settings here. If you want to change the variation of the image, or if you want a variety of this image, you can just change the denuance section here. Okay. Now after that, you simply have to go to prompt. That's all. Once you Q prompt it, all the things or all the nodes here will not affect it. As you can see, VAE decode here is red. So this node system won't work. The only thing work here is that this upscale section only. And you can see ultimate SD upscale is now working here. We have to wait for some time because upscaling might take time. Once the preview is now, as you can see, as I was going to explain it to you that once preview is here, you can see your image is getting scaled up. The time is depend on the size of the image. If you want to scale it by fax, you can do so and it will take time, depending on the power of your graphic card or machine. It simply scale it by this factor. Five, one, two X five, one, two. It is scaled like a tile section. First, it create the upper right, then first, it scale up the upper left, then upper right, then upper, then lower left, lower right, one, two, three, four. One, two, three, four. This is how it scale up. As you can see, it is now working on the upper side sex. This is the section, upper right. And after that, it will work on lower left then lower right. So here is the upscale image, as you can see, I only make the background is black for the PNG file. So it's saved in our folder, that is the output folder of CPUI. We have to go to Comp ui folder in the CPI in which you have installed it. Check for the output folder, and here is your file double in size. That's all. You can upscale it even further. Depend on your use of the image. As you can see, the size is 2048. The original size was 1024, so it doubled the size of the image. That's odd, as you can see. Now here is the upscale. You can use it anytime. Without any changes or without creating any of the image. So this is the file that you can find in the resource section. Thank you for watching. See you in the next lecture. And I really appreciate that you please give it a five star rating. It keeps me motivated for my new lecture. See you again. Bye bye. 51. Create Anime images using loras from Civit AI: Any five star ratings I'm receiving to this course, and I believe that you love this course. So we are going to move forward. We are going to use some extra as to generate our images according to us. Now, let's use a workflow that I have already shared with you. So this is the workflow that we are going to use workflow for and once you drag this workflow here, you will see this is skin pop up. And if there is any problem or there are node is not available, that might not happen. According to me, I guess, there will be no problem with this workflow as of now. If you get any problem in this workflow, you can message me directly or start a discussion in this course section. Anytime. I will help you out. So if you have any kind of problem you are facing, you can ask me directly and check for this section, Section number 11, discussion of ask. I will update a video here so you can check it here if you have a question. I will just record a video so that other can understand that thing I was explaining to you specifically. So it will help everyone with the same problem. So if you have a problem, you can ask me anytime I'm here for you. So right now I'm at civiti.com. I simply sign in Civ AI. So right now we are going to work with these kind of images. Let's start with this. I think we call it Wifu Takari Saske, nub AI, Laura seven, checkpoint. We are not going to use this one. Let's check it. Net pus Civ cha. So this is CVCa and we are going to use this Laura for ourselves. Right now, just click it. And remember, they have a triggered word here. These are the triggered word training images that you use to train this Laura. And right now, let's simply download this Laura. It is around training data, Laura. Let's download this Laura. Where you have to save it, you simply have to go download in the CFUIPortable window where you have already installed a CPUIopen Comp UI, check for models, search for models. This is Double click, Laura, ChiPcha save and they have a triggered bud. Let's check it. So you have already downloaded the aura in your folder and they have a triggered word, and I'm pasting it in our Google Docs. I will share the link with you so you don't have to worry about that. Copy, paste, comma Copy, Paste. So here are the three tigaword that they are using. Creating a doc for trigger words is important because obviously we will forget this word. So we have to copy these words in our doc. So once you get these Tigabats I will simply copy the link of this Laura and paste it here so you can use it directly. And I also copy the title of this Laura. So you will understand it easily. Paste it. So here is the copy of this. So now, once you get it done, simply go to your CPUI. You have to go to At refresh not definition. That's all. Once you refresh it, it will take some time, about 1 minute or 2 minutes depending on the power of your screen. Oh, I just happened in seconds with me. So once you get it done, here is a a name will appear here. Say, ha, here is the a. Click it now, Sivcha is appeared here and there is no Laura has been selected here. Flux tape, I'm using Snell. Why am using Snell? Because it has steps, which is lower. So I am using four steps for this one. And as the upscaler is already off, and you can upscale anytime, right? We have discussed it in our previous section. In our discussion in our help section, you can check that how we can upscale image. So once you Q prompt it, let's simply copy these three Tigaworts, copy it and now paste it here. But I want help from Chat JPT create a prompt for scene of UW Girl with text, I will find you. Let's enter it. Chat GPT will give a prompt for us. Let's copy it and paste it in our prompt clip text and code. Once you copy it, you have to copy the trigger word also, copy and paste it in our Chip. Now, press Q prompt. Now wait for some time. First time it will take some time to load the loa or load the model. Loading the model might take some time. We want it like this one, right? The one I can see here. So to understand that, let's click on one of the images. After that, check photo prorompt if there is any. No. And if there is no update here, let's cute anime style wi fu girl standing. Copy, Paste, shicha a soft pink wearing a pastel hoodie. No, let's move this wearing with cat ears. With cat ears holding a glowing phone, her eyes spike cle subtypukO lips neon hurt float around it in a poldTreamy letter above her head, the text reads, I will find you. The scene is a dreamy wp her wb starting with sparkle, soft lightning and plus tone. She looks adorably determined yet playful ShihaRmove this. Let's Q prompt it again. Let's specifically work with this type of generation art. And for this one, let's use this prompt. Show more prompt, copy, and we have a prompt here. As we have already discussed that negative prompt won't work in Flux. So let's simply Q prompt it and understand how these things work. Master work, masterpiece, past quality, hyper detail eight k, USD, one girl, Civcha pink hair blue eye. Made uniform. Okay, that is the trigger we I think made uniform, Laura IVa 0.09 quality masterpiece, extremely aesthetic, high detail dynamic composition, vibrant color, vivid color. So they have not explained a complete type of scene, right? I will find you. This is kind of looking dreamy. And yes, we are getting results like this one, right? But yeah, image generation is good. Which type of generation model they have used? Guidance, Step 35 A. This is the metadata that they have used, and if we use steps till 35, yes our generation will take more time. Results are good, not bad, that word w and let's understand. Modal bit, Laura, why don't we just simply paste the trigger word here. Let's work with that. Let's work that type also. Once we click only the trigger word, let's see what we get here. SiphaPope pink here. Now let's work with Deb and increase the steps two, I guess 20. Let's make it 20 and they have a Euler sampleym Euler and guidance is 3.25 0.5 that they have used. They have a seed number. I think they have also a seed number for that. Seeds this one is specifically Euler a 35. See this paste and steps 35. Let's use this specifically, and I hope this time we get the desired result. Let's prompt it again. Before that, let's copy this page. And if there is anything they have used here, only they have used this data generation data, solo, tag, and what they have used, nothing. Let's you prompt it. And this will take some time because we have 35 steps to process. Let's jump to the future again. Results are pretty good, but not like this one. Click it. Let's use this prompt. And paste, paste it, Q prompt it. Let's wait for some time again. And let's see if we get the result here. Results are not like that, but yet, it's kind of cute and it's kind of not cute. And these are the killer eyes, I think, and these are the cute eyes. Now, let's keep it to Snell version and the steps, keep it four steps for better and faster performance, keep it 3.5. 52. How to install and Create images using Flux redux: Simply copy this workflow here so you can find it by clicking here or you can find it in the resource panel. I have shared the workflow here from redux style mix up flux to Images workflow. I simply downloaded it and once you download it, you have to extract it and after extracting, there is a file. Here is the file, and you simply have to drag this workflow here in your CUI. Once you click Area, once you drag the workflow here, there will be a alert open here. You simply have to press X once you have done it, now you have to go to your ComfUI manager, Install custom machine node, install update all. That's all you have to do. Now, it might take some time. Meanwhile, you have to download these files that I have shared here. This is a link. Once you copy this link and paste it into here, it will directly download your file. And to download this redux file, you first have to log into huggingpas because it might not give you the access to download it. So you have to go to your hugging phase and copy this text here. Copy is restricted. Go to hugging phase. In the search panel, you have to paste this text here, enter to access this file to download this file, you would simply have to go to files and Verson and there might be and there will be a access permission, they will ask you to download this file once you accept that and give them access to control to download the files. After that, you can simply go to your workflow and copy this and paste Once you give the access, it will start downloading directly. Here is the workflow. You can directly download it. And to install it, you have to go to your CPI folder where you have install it, go to models, and go to style model. This is the folder. And you have to download Flux dux here. I have already downloaded it, so that is why I'm not saving it here. So I'm canceling it out. And now, si clip, you have to download this into CPI model, clip Vision folder. In the models, clip Vision folder, you have to download it and save it file and save the file there. Now, basic setup is done. Now let's check for second node. The redux model lets you prompt with images and we use flux tab and Snail model workflow. We have already downloaded it. You can chain multiple applies style models node if you mix multiple images together, if you get error of five in the correct direction, see the top of the example of the page link, flux fave, Comp UI. So you have to download fluxdavmdel dot saf N in CPI Models Unit. Let's check it out if we have done it already or not. For that purpose, let's go to Comfy Window CompuI and models select for UNT. In the Unit Seton, flux Dave fun. Yes, we have already did that. And if you have not done that, you have to download Flux Dave model in the Unit folder. And I believe if you have followed my lectures already, you have done it already. So for that purpose, let's check for T five XL FT Safner into CPI models clip. Now, go to CPI Models and check for clips. Once you check for it, TTX F 16, yes, we have already done that. So we don't have to worry about anything. Now, AE safeneres in A E safener goes in model VA E. Now go to models and check for VAE, VAE. Here is a VA folder. And yes, Asafn we have already got it. Now, you can set the weight deeper above the top if memory issue. Okay. Now, you can control, weight standoff images implies style mode. For using GGF model, right, click on G group node, side group node to always, uf install machine custom node or direct from Comi Manager. Unit loader, Comfy sampling flux, Duallip loader, flux Dave UFT M these are the models that you have to download to use GGUF Model loader that they are referring to. But right now, we are not using it. And if you are using it, you simply have to connect this model to here and this clip to here. That's all you have to do. You simply have to replace the images. That's all. Now, line in vector art style. Now we have to upload the images. So these are the images that we have already generated. Let's use this image open and now choose another image. Select it open. So these both are the images that we are going to redux. Let's Q prompt it, and I hope this will work. And if not, we will rectify it. So here is the problem that we have. Prompt output failed validation VE loader. Value nought list, In A is safener not in flux, A is safer, Sefner VAE. Unit loader, Value naught is list unit name flux Dave one. So close it. So here is the flux tape at. Let's check it. Yes. Now you can check for the AEVloder now check for the VAE loader. Where is the VAE loader? Crop center crop center clip vision. I have to check for the VAE here is the VAE loader. A Effner, here is a AE for flux, for flux AE, let's check for Q prompt again. Yes, now it's working. So here's the so problem here is that our file name is not match with the file me that they are using in the workflow. That is the problem, right? Now, and I hope we will get something amazing. Let's close our eyes and open after 1 minute. Wow. We got the result. Wow, we got the result. It looks not that bad, actually. Now, let's use image. So we are using image of Naruto. And after that and now use and let's use Gso here. Now, now let's Q prompt it. Now we are mixing Naruto and gozo, and if you watch anime, you already know it. If we are getting something, let's change this image to Goofy AD image. The save image, save. Now let's check it out. If we get something. After that, prompt it again. So I have used this is the image that I have generated. This is the image that I have downloaded from Google. And after this generation, we will just Qu prompt it and wait for the generation also. Why waste time? Let's do it. Imagination is not that bad, but yes, the quality is really bad here. Okay? They are taking the collector together, and we have ofi. I think the quality that I have used before is low quality. That is why the quality that I have received here is really bad also. And right now, I can see the sharpness in my images. And here I can check. After the generation, let's see what we will get here. So here is the generation we have received. Not that bad, actually. Let's check our output image of CPU y. Here's the copy output, and here's the image. So that is how reducts work. It simply makes two images and give you mix two images and give you the output. That's how image reduct work. And see you in the next lecture. Thank you for watching and thank you for giving this course five star rating. I really appreciate, and it really motivated me to create lectures further and further and further. See you again. Thank you for watching. 53. Flux Redux overview and use mix 2 images and create something new: Simply copy this workflow here so you can find it by clicking here or you can find it in the resource panel. I have shared the workflow here from reduct style mixer flux to images workflow. I simply downloaded it and once you download it, you have to extract it and after extracting, there is a file. Here is the file, and you simply have to drag this workflow here in your CPU. Once you click Area, once you drag the workflow here, there will be alert open here. You simply have to press X once you have done it, now you have to go to your CompuI manager, install custom machine node, install, update all. That's all you have to now it might take some time. Meanwhile, you have to download these files that they have shared here. This is a link. Once you copy this link and paste it into here, it will directly download your file. And to download this redux file, you first have to log into hugging page because it might not give you the access to download it. So you have to go to your hugging page and copy this text here. Copy is restricted. Go to hugging page. In the search panel, you have to paste this text here and to access this file to download this file, you simply have to go to files and erin, and there might be and there will be a access permission, they will ask you to download this file once you accept that and give them access to control to download the files. After that, you can simply go to your workflow and copy this and paste here. Once you give the access, it will start downloading directly. Here is the workflow. You can directly download it. And to install it, you have to go to your CPI folder where you have installed it, go to models, and go to style models. This is the fold. And you have to download flux dux here. Okay? I have already downloaded it, so that is why I'm not saving it here, so I'm canceling it out. And now, SiClip you have to download this into CUI model, clip Vision folder. In the models, clip Vision folder, you have to download it and save it file and save the file there. Now, basic setup is done. Now let's check for second node. The redux model lets you prompt with image and we use flux tap and style model workflow. We have already downloaded it. You can chain multiple applies style models node if you mix multiple images together, if you get arr of files in the correct direction, see the top of the example of the page link, flux one tape, CPUI. You have to download flux dapmdel dot saf N in CPI Models Unit. Let's check it out if we have done it already or not. For that purpose, let's go to Cofi Window CPUI and Models select for UT in the Unit SeconFlux Dave, we have already did that. And if you have not done that, you have to download Flux Dave model in the Unit folder. And I believe if you have followed my lectures already, you have done it already. So for that purpose, let's check for T five XL FT safener into CPI models clip. Now, go to CPI Models and check for clips. Once you check for TTX F 16, yes, we have already done that. So we don't have to worry about anything. Now, A Shaffner goes in, A E Sapner goes in model VA E. Now go to models and check for VAE, VAE. Here is the VA folder. And yes, A Sapner we have already got it. Now, you can set the weight above the top if memory issue. Okay. Now, you can control weight standoff images imply style mode. For using GGF model, click on Grookne side group node to always. You have install missing custom node or direct from Copy Manager, Unit Loader, Comfy sampling flux, dual clip loader, flux DPF TM these are the models that you have to download to use CGUF Model loader that they are referring to. But right now, we are not using it, and if you are using it, you simply have to connect this model to here and this clip to here. That's all you have to do. You simply have to replace the images. That's all. Now, line in vector art style. Now we have to upload the images. So these are the images that you have already generated. Let's use this image open and now choose another image. Select it open. So these both are the images that we are going to redux. Let's Q prompt it, and I hope this will work, and if not, we will rectify it. So here's the problem that we have. Prompt output failed validation, VE loader. Value list, V A is safe ner, not in flux, A is safer, safener VAE. Unit loader, Value naught is list unit name flux Dave B. So close it. So here is the flux tape a eight. Let's check it. Yes. Now you can check for the AEVloder now check for the VAE loader. Where is the VAE loader? Crop center crop Cena clip son. I have to check for the VAE here is the VAE loader, AI Sefner here is AI For flux for flux AE, let's check for Q Prompt again. Yes, now it's working. So here's the problem here is that our file name is not match with the file name that they are using in the workflow. That is the problem, right? Now, I hope we will get something amazing. Let's close our eyes and open after 1 minute. Wow. We got 30, sorry. Wow, we got the result. It looks not that bad, actually. Now, let's use image. So we are using the image of Naruto. And after that and now use and let's use Gso here. Now, now let's Q prompt it. Now we are mixing Naruto and Gozo and if you watch anime, you already know it. If you are chatting something, let's change this image to fi AD image, then save image save. Now, let's check it out. If we get something. After that, que prompt it again. So I have used this is the image that I have generated. This is the image that I have downloaded from Google. And after this generation, we will just que prompt it and wait for the generation also. Why waste time? Let's do it. Imagination is not that bad, but, yes, the quality is really bad here. Okay? They are taking the arrector together, and we have ofi. And I think the quality that I've used before is low quality. That is why the quality that I have received here is really bad also. And right now, I can see the sharpness in my images. And here I can check. After the generation, let's see what we will get here. So here is the generation we have received. Not that bad actually. Let's check our output image of CPUI here's the CPI output, and here's the image. So that is how reducts work. It simply makes two images and mix two images and give you the output. That's how image reducts work. And see you in the next lecture. Thank you for watching and thank you for giving this course five star ratings. I really appreciate, and it really motivated me to create lectures further and further and further. See you again. Thank you for watching. 54. Flux Redux using GGUF Model: Welcome back, everyone. So we are going to use another model that is other than flux, and this is GGUF. And if you have followed my course thoroughly, then you already know about this GGUF model. Now, let's Control M. I think Control M is not working. Let's right click. Nod go to next node. Workflow Image, edit group. Yes. Let's select this one. Not the group. I don't know why, but directly, I can do this thing with a group A. But now let's select for the model. Here is the model. And once you click it, select this model, q five before there is a qu eight model, and just change it to this and for clip, let's change it to this. Everything will be same only this thing has been changed. So what we have to do We have to run the prompt. Well not list in the TE. Okay. For this one, we have to change this also. Let's scrape TT XXL. Let's run it. So I simply what I did, I simply changed the model that so have done. And to increase the size of this, we can change the width and height of this image. We will do with the next run up and with the update, we can see here is a run up. And here we will experiment with this value, once we're done with this one. So here is a result, somewhat same. I don't think there is so much change in that. Let's change it to Naruto. I think we have to change it to Naruto. I simply download all this image from Internet, so you can experiment it out. So let's strength it. First, let's experiment it with this particular value. After that, we will work with the strength and multiply things. Not that much good. Let's keep the strength to five and change the antibs strength type. Let's make it run again. But yeah, it's taking the shape of the image like this one. So here is another image, we can see here, here. So let's change the strength of Naruto to one and change the strength of Lufi 0.7. Let's run it again. You only have to experiment with these things. The value of these image, you have to experiment them. And I think right now we are getting the result, which is really fine. And the results are acceptable this time. Yes, now we have LFI with Naruto eyes. Yes, it's at least acceptable, better than before. Decrease the stngth further, we will get another result. So you get the idea how we can use the redux model with GGF model and create an output of our image. Now let's create I download this monster image from Internet. You can download it from anywhere. I simply write monster PNG and download it. And after that, there is a scene of city So this is a random image with people. I simply download this image. I'm taking a random picture from Internet. So I just simply put this image and with the strength of 10.7. So for this image, we want the image of this one is one because we want this image to work with this one. And yes, we want strength of this one. Let's connect both the images. We want a 0.8 because we want this thing to be printed on this one. Okay, so to do that, let's experiment with this value. After that, we will interchange the value. So you will get the idea better. Now let's run it again. And wait for some time. I think for some reason, the GGF model is working fine. I can see monster here. And now we have the Awesome. It's awesome. I love this image. How this beautifully changed this into this. I really love this image. How this thing changed this to this. Really loved it. Whoa.