Higher resolution pretrained model ? #188

Open
opened 2026-01-29 21:45:29 +00:00 by claunia · 3 comments
Owner

Originally created by @Michaelwhite34 on GitHub (May 13, 2022).

512512 is generally enough for video, but portrait needs 10241024 or 2048*2048.

Originally created by @Michaelwhite34 on GitHub (May 13, 2022). 512*512 is generally enough for video, but portrait needs 1024*1024 or 2048*2048.
Author
Owner

@johnGettings commented on GitHub (May 18, 2022):

Check out the --upscale argument. Default is 2 so your image is initially 256x256. Change that to 4 and output will be 1024x1024. Results are great too. It utilizes RealESRGAN

@johnGettings commented on GitHub (May 18, 2022): Check out the --upscale argument. Default is 2 so your image is initially 256x256. Change that to 4 and output will be 1024x1024. Results are great too. It utilizes RealESRGAN
Author
Owner

@Michaelwhite34 commented on GitHub (May 18, 2022):

Check out the --upscale argument. Default is 2 so your image is initially 256x256. Change that to 4 and output will be 1024x1024. Results are great too. It utilizes RealESRGAN

I am talking about enhanced face resolution, not the enhanced background resolution.

@Michaelwhite34 commented on GitHub (May 18, 2022): > Check out the --upscale argument. Default is 2 so your image is initially 256x256. Change that to 4 and output will be 1024x1024. Results are great too. It utilizes RealESRGAN I am talking about enhanced face resolution, not the enhanced background resolution.
Author
Owner

@johnGettings commented on GitHub (May 18, 2022):

Just try --upscale 4. It's what you're looking for. I think they use realESRGAN for the background and something else for the face. I don't remember. It comes out 1024x1024 and it looks good

@johnGettings commented on GitHub (May 18, 2022): Just try --upscale 4. It's what you're looking for. I think they use realESRGAN for the background and something else for the face. I don't remember. It comes out 1024x1024 and it looks good
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: TencentARC/GFPGAN#188