Comfyui Advanced Tutorial 03 - Flux Kontext

0:00 Welcome to the new episode. Today we talk about flux context, image variations and image editing. Time to start. We load image here. After that we have load model nodes. This minus zone node is important. It allows you to load model partially so that you can use workflow at 6GB VRAM. Loa loaders which is disabled not working. Flux context scale. It converts image size to lower resolution so that model can work. your prompt here. Then we come most critical part. This primitive controls down sampling. It adjusts how the image will change. Close to value one makes familiar to original. Close to five is more prompt following. I gave you notes so you should read that. We will use it

0:47 on five mostly. Also, I have high- res fix here but not stable or efficient yet. So, I am not releasing it for now. Time to get into the prompts and results. Part one, change weather, time, color. Okay, the prompt is changing the car color, add custom car wrap, weather, car adjustments, and time all at once at the same time. Hard job. And it really did. As you remember, without Redux, it couldn't do it. This workflow is not only increasing the consistency, but also increase the prompt following. Another one. This one also has people and car behind. It is all consistent but only quality decreases. To solve it, we

1:34 need TTP with low D noise. Part two using this style prompt. I think this is the one you will like most. Now we can make our thumbnail maker. Not only thumbnail but get reference from some image and make same style without prompt. We use primitive 4.5 for this. Lower for more consistency. As you can see it is too consistent. Even the face is same if we want. You start prompt with using this style. After that you describe how you want your image. Then you get same style plus you upscale it. You don't even need to make thumbnail anymore. I will give you all the prompts used in this video in files so you can try it. Part three. GPT4 type prompts. I think

2:25 you saw them online like the object made from glass or made from wood etc. I also tried it and it is working. I told Gemini to give the prompt that this cap is from glass and wood version. It worked. If you want to try last one is even better. I prompt the car photo to be made from wool. It is very cool. Maybe we should even try to make it like Hot Wheels. Part four, change image style too. The prompt is simple. For example, change image style to cartoon style, Minecraft style, pixel art, anime, and more. My favorite one is pixel art. Should be tried cinematic, maybe even with film names. You should try it. Endless

3:13 possibilities. Part five, changing the text and close. We are now getting close to image editing part. To change the text, you should make the prompt like change the text from text one to text two. You should tell full text on the prompt as I made result is good. Although we couldn't copy the style of the text, but we will. Don't worry. On the inpainting part, changing clothes is also easy. You can tell like change outfit too. And to add text, you can say like add a text called text one with the text properties. Maybe even font should try. Now it is time for inpainting part. Here

4:01 we will mask the area we want to change and keep the quality same without decreasing quality. First you should open mask editor and mask the area you want to change. Then type your prompt such as adding object removing object changing color. Notice that the output is same resolution as input only the masked area changed. Let me show you different examples with prompts. Remove any object. Just tell what your object is. If you can't identify like it is weird object, you can tell like brown color object. Even very tiny object can be removed as I told you before to make your image realistic. Also changing the color. Now I made prompt couch color but I did not mask the pillow so it stayed

4:47 same. Lastly I added magazine book with my properties on the table. It worked. So you can add objects too with properties. That is it for today's video. Next video will be about object migration or depth map with high-res fix. I need to decide. Thank you for watching.