search

AUTOMATIC1111 / stable-diffusion-webui

Stable Diffusion web UI
GNU Affero General Public License v3.0
139.14k stars 26.41k forks source link

Use zoomed in Stable Diffusion for face restoration #2125

Open lendrick opened 1 year ago

lendrick commented 1 year ago

Is your feature request related to a problem? Please describe. The existing face restoration models work well on photos, but not on cartoons and anime. Right now, I can improve anime faces by manually going to inpainting, masking the face, turning on inpainting at full resolution, and setting the masked content to "original", and this works extremely well, but it's a manual process. Since it's already possible to detect faces, I feel like it ought to be possible to run Stable Diffusion with the same prompt (or optionally with a different one) zoomed in on the face and then resize it the way that can be done manually with inpainting.

Describe the solution you'd like Allow us to select SD as a face restoration option, and then do the above automatically.

Describe alternatives you've considered Doing it manually.

Additional context Add any other context or screenshots about the feature request here.

Ehplodor commented 1 year ago

Could be added as a fixer in the future (maybe) "fixer tab" ? Asked in https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/2162

lendrick commented 1 year ago

It could, but it would also be nice if I could mass-generate images without having to take each individual image to the fixer tab to make the faces look decent.

Ehplodor commented 1 year ago

It could, but it would also be nice if I could mass-generate images without having to take each individual image to the fixer tab to make the faces look decent.

Batch option in Extra tab ? Idk if it is already implemented.

Seems that separate fixer tab and scaler tab is not on the way...