Face Restore (GFPGANv1.4.pth) is not effective for some images

[QunBB]

First, confirm

• I have read the instruction carefully
• I have searched the existing issues
• I have updated the extension to the latest version

What happened?

As the following figure, you could see the two output faces are both blurry. One is with face_restore_model, another is not.
It looks like the face restore is not effective.

Steps to reproduce the problem

Sysinfo

Ubuntu 24.04.1 LTS
Chrome browser
NVIDIA GeForce RTX 4090
torch 2.4.0+cu124

Relevant console log

got prompt
[ReActor] 11:55:49 - STATUS - Working: source face index [0], target face index [0]
[ReActor] 11:55:49 - STATUS - Analyzing Source Image...
[ReActor] 11:55:50 - STATUS - Analyzing Target Image...
[ReActor] 11:55:50 - STATUS - Swapping...
[ReActor] 11:55:50 - STATUS - --Done!--
[ReActor] 11:55:50 - STATUS - Working: source face index [0], target face index [0]
[ReActor] 11:55:50 - STATUS - Using Hashed Source Face(s) Model...
[ReActor] 11:55:50 - STATUS - Using Hashed Target Face(s) Model...
[ReActor] 11:55:50 - STATUS - Swapping...
[ReActor] 11:55:50 - STATUS - --Done!--
[ReActor] 11:55:50 - STATUS - Restoring with GFPGANv1.4.pth | Face Size is set to 512
Warning torch.load doesn't support weights_only on this pytorch version, loading unsafely.
Prompt executed in 1.64 seconds

Additional information

No response

[darebgd]

Defo a bug. Give that girl the same haircut and there you have it!
P.S. Try with fully formed faces, maybe it will work better 🤣
It's face swap, not head swap 😮‍💨