Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Some ideas to improve quality #39

Open
2blackbar opened this issue Apr 4, 2023 · 2 comments
Open

Some ideas to improve quality #39

2blackbar opened this issue Apr 4, 2023 · 2 comments

Comments

@2blackbar
Copy link

2blackbar commented Apr 4, 2023

SD generates best face and likeness when its close to native res or higher and degrades faces when res is below 480 (easy to test)so...

  • Do You guys upscale the face to 512x512 or whatever resolution is set in img2img ? Even if its low res face that has like 200x200 pixels , upscaling it to 512 or 768 would improve the quality, It would be a good idea to tinker with .

  • Another thing that helps with low res images is sharpening, what i do wheni have lowres image is i upscale it until face size is about 400-500 pixels, then i do sharpening so its a bit crispier detail, and then i inpaint the face , this really makes a difference , sharpening tick box maybe with a scale of how much , like 0.2, 0.3 would be nice to automate all this so i dont have to upscale on my own before inpainting the face .

  • Even better if You would be able to detect the face and also detect if its resolution is lower than 512, and iif it is, resize it so its 512 pixels wide.

@2blackbar 2blackbar changed the title Some ideas Some ideas to improve quality Apr 4, 2023
@kex0
Copy link
Owner

kex0 commented Apr 5, 2023

Hey, thanks for the suggestions.

Currently it generates the face at the resolution you set in the webUI and then resizes it up or down depending on the size of the original image.

The sharpening is an interesting idea, I might look into it.

@2blackbar
Copy link
Author

2blackbar commented Apr 6, 2023

IT would be worth to test if resizing the face to lets say chosen 512x512 before passing to img2img denoise would improve the results instead of giving it smaller image at original size and img2img does upscaling to 512x512, from my tests it improves a lot when you do upscaled before img2img pass so img2img works on 512x512 or close resolution from get go , and even better if its sharpened

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants