How To Do img2img Within ComfyUI - Beginner's Guide
ฝัง
- เผยแพร่เมื่อ 23 ก.ค. 2024
- #stablediffusion #aiart #generativeart #aitools #comfyui
As the name suggests, img2img takes an image as an input, passes it to a diffusion model, and outputs an image. Here's how to do this process within ComfyUI.
Time Stamps
Intro: 0:00
Setting Up the Nodes: 0:11
Modifying the Prompt: 1:40
Decoding the Image: 2:39
Processing the Image: 3:30
Outro: 3:55
Workflow:
promptingpixels.com/comfyui-w...
Important links:
- 👾 Discord: / discord
- 🌐 Website: promptingpixels.com/
- 🛠️ GitHub: github.com/content-and-code - วิทยาศาสตร์และเทคโนโลยี
I've watched dozens of comfyui tutorials. You're the best at explaining every step without rambling, or pretending we already know the basics. Gonna watch everything you got in this channel :)
Thank you for the nice feedback - happy to see the tutorial was useful for ya!
Perfect and simple explanation ❤🎉 thank you.
Straight to the point and easy to understand~ Thanks for that.
Very nice explanation.
Thanks for your amazing tutorials! Very straight forward and easy to follow! I've spent days looking for a tutorial that was easy to understand and actually worked! You're definitely the best! I'm a newbie, do you have more tutorials on what each node represents and how to download / setup models, loras, controlnet etc.? And also can we run ComfyUI in Google Colab Pro? Thanks, again!
Perfect video
thank you so much
Very good, simple, straight to the point. Thank you sir. I have a doubt - Yes I want to use a reference image, but also want to change the aspect ratio, size and resolution for the output. As we cannot connect empty latent node directly to ksampler now, how do i do it?
Can you use img2img like midjourney, where an image is used as "inspiration" for the output image but the whole image changes instead of bring the same image/background/position?
great tutorial. thanks buddy. Please do more...
Absolutely - anything in particular you'd like to learn more about?
Yep. text writing or text animation on AI Videos. Could you please make a video about it ?@@PromptingPixels
Thank you for the suggestion - i will look into this and see what I can wrangle. Text is always pretty tough with diffusion models, so there might be some limitations here. Stay tuned.
Thanks@@PromptingPixels
Hey - just remembered this comment. The Stable Cascade does rather good, though not reliably, text writing in the outputs. May be worth checking out for your project.
Can you try the image-generating process with an unclear photo? Many times AI cannot perfectly represent the details and materials of the original image, such as the style and material of the clothes worn by the clothing model.
Is there a way to encode an image, but scale/crop it down to a desired aspect ratio in comfy UI?
Thanks so much!! This was a clear explanation of img2img. What if I have a pic of myself? I wouldn't have a prompt to start...
Sure you would! 1man, smiling, glasses, etc. - just describe what you see in the picture and that is your prompt :-)
oh wow...will definitely be testing that. @@PromptingPixels
what if i want to change a real life photo to stylized while keeping the model face and pose same?? example : changing the model's clothes to be futuristic or change the background into a forest
Pose you'll want to use ControlNet. For face, either a specially trained LoRA, face swap, IP Adapter (it won't retain likeness but can help with general features). The rest would be prompting.
Большое спасибо за подсказку. У меня получалось изображение не имеющее связи с образцом, оказалось что я забыл уменьшить шум с 1 до 0,5. Теперь все идет как нужно.
Thanks a lot for the hint. Earlier, I received an image that has nothing to do with the sample, it turned out that I forgot to reduce the noise from 1 to 0.5. Now everything is going as it should.
Can img2img do which the img is Pop Art or Abstract Art?
Yeah definitely. Easy to do with landscapes, scenes, etc. as you can raise the denoise (probably around .8-.9) and change your prompt accordingly. However, if doing this to characters, people, etc. you'll lose some of their likeness in the process. To counteract, I think using an IP Adapter or LoRA may help to counteract this.