0:00 Welcome to another tutorial. Today's concept is consistent characters. More advanced editing presentation of a workflow I called context image gen. An investigation of can we make our consistent character image without Laura training or at least getting variation of our character. These are good questions. Time to start with context editor v2. I assume that you watched the first video of context. So what is different here? Well, first thing is I noticed that if you change the seed, it makes completely different image related to the parameters I have given you. Our prompt is changing the car color lights and add a text on the ground. You should lower the redux related to this seed to
0:47 not making completely different image. Result is good. Now let's see what improvements do we have. The high-risisk fix added it adds details, remove the blur a bit and higher resolution. Only problem is it can ruin the text sometimes. So be aware of it. You can easily see that resolution is higher. It corrects the inconsistent details if you have. Now let me show you the context image gen which is being used for only making different images with flux dev lauras high resolution and different concepts. We use two different things here. First is Laura. Second is we make prompt like fluxdev workflows very long and detailed prompt. The prompt adherence is so high in context. So we use it for this image generation. I will
1:33 also give the Laura I am using in this workflow. Also you can choose a concept. On the right we have high restix. I strongly recommend to wait and see how this works. It adds details and try to fix related to the prompt. Now you may ask this question. Can't we make the same image with official workflow they brought? Sadly impossible. Compare them and look at the prompt. Which would you choose as a winner? I will leave all the prompts on free patreon as you know the winner is obvious I guess. Now time to see the consistent character project. The workflow name is flux context plus pulid. I just write long and detailed explanation on the left. So you can read that and also prompt structure examples under the explanation. So first we load
2:20 face image but how far the face will be away from the camera on the reference. The closer you put, the more detailed face will be in the output. But do not make it too close. Make it like medium portrait. And make long detailed prompt. Make it close portrait or medium portrait. Do not try to make full body. We are making face consistent here. So this is our first output very familiar face. The resolution is low. So we need to make more detailed with high-risis. By the way, I think you have pulled installed already. If you don't have you cannot install it with just comfy UI manager. You must install face xlib and other things. If you get import error I will give you a video link to install if you are new on this. After you wait the high-risisk fix you will get this output
3:06 more humanlike details and higher face resolution and consistency. This opens us a window to make our character with different environments. Now the last workflow name is context plus pulid face editor. In this workflow, we will only change the face expressions of humans to get variation of our image. You can control one face or even two face with this. The logic is same. We use pulid to make consistent change of expression. The workflow is almost same as context pulled. The difference is we do not use Laura here. You can try if you want to, but it can change image. The prompt structure is like this. Rotate face, make her smile. On the right side again we have hyris fix detailer. So total of
3:51 four workflows I have given you. Now lastly we see some examples with some different lures and settings. Now time to see some examples. First one is concept art. You know who he is. This image made with context editor v2 workflow. I will give all the prompts loras and settings on this workflow. You can easily see that hyros fix worked well and fixed the missing text on the photo. I did not do anything. It fixed itself. The second image is made with context imaging. Einstein's quote with many texts on the prompt like a magazine cover. It is good. Iris just did good and fixed some text as you can see. Now context plus pulid workflow. We put
4:39 input as face and get different environments with consistency. I just showed you the first pass outputs. If you wait for high-risisk fix, you will get much better resolution. And lastly, some more context editor outputs. This is a 3D character with different environments. Redux values are low depending on the noise seed. You can see it is consistency. I don't know how many times I said the word consistency, but here we are. Thank you for watching this video. Hopefully meet with the new one.
