mirror of
https://github.com/TencentARC/GFPGAN.git
synced 2026-04-23 22:30:43 +00:00
Black Background, with enhanced face #198
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @AntonHarboe on GitHub (Jun 27, 2022).
Hi, i have been having this issue where i upscale the picture, then when i look in my output folder, there is just pictures with a black background with the face on the specific spot it should be. pls help
@gerrith commented on GitHub (Jul 31, 2022):
The same here:
Not sure what I've changed, the day before it was working ok. I think it started after I've installed Real-ESRGAN.
@fogostudio commented on GitHub (Aug 24, 2022):
same problem
@GordonFreeeman commented on GitHub (Aug 30, 2022):
Anyone managed to find a solution? I'm having the same issue, but no Real-ESRGAN installed. I had a problem with blank green screen generation in Stable Diffusion, that was related to fp16 not working on my 1660 Ti, could this be something similar? And if so, does anyone know how to switch to fp32 (if this is even relevant at all)?

@differentprogramming commented on GitHub (Sep 7, 2022):
same
@RupertAvery commented on GitHub (Sep 10, 2022):
I got this problem too after setting up my environment with CUDA "properly" so that torch detected it. I have a GTX 1650 with 4GB VRAM.
I was able to fix this by editing
inference_gfpgan.pyand settinghalf=Falsewith RealESRGANer, so it is probably like the issue with --fp32 in SD.@cgessai commented on GitHub (Sep 12, 2022):
@RupertAvery Worked perfectly on my GTX 1650- I thought I'd installed pytorch incorrectly. Thank you SO much!
@GordonFreeeman commented on GitHub (Sep 12, 2022):
That did it, thank you very much! 🙏
@ZahirAM commented on GitHub (Nov 14, 2023):
Got the same problem with only black spot on the face
UPDATE- Solved by putting down the image from 16-bit to 8 bits/channel
@HexNIK commented on GitHub (Dec 17, 2025):
@RupertAvery
You've been a great help — thank you from the bottom of my heart.
p.s. NVIDIA GeForce GTX 1660 Ti, CUDA Version: 12.4