[Bug]: Generate image was corrupted

[naoyaikeda]

Checklist

• The issue has not been resolved by following the troubleshooting guide
• The issue exists on a clean installation of Fooocus
• The issue exists in the current version of Fooocus
• The issue has not been reported before recently
• The issue has been reported before but has not been fixed yet

What happened?

I create a prompt after generation image was corrupted.

I had heard of a similar problem with a1111, so I assumed there would be a similar situation with fooocus.
The code I suspect is the implementation of patched_encode_token_weights in patch_clip.py.

 z = z * (original_mean / new_mean)

In the above line, it seems that there are cases where the new_mean is much smaller than the original_mean. In fact, if we dump these values,
original: tensor(-0.0196, device='cuda:0')
new: tensor(-0.0199, device='cuda:0')
original: tensor(0.0005, device='cuda:0')
new: tensor(0.0001, device='cuda:0')
original: tensor(-0.0159, device='cuda:0')
new: tensor(-0.0159, device='cuda:0')
original: tensor(0.0079, device='cuda:0')
new: tensor(0.0080, device='cuda:0')
original: tensor(-0.0202, device='cuda:0')
new: tensor(-0.0205, device='cuda:0')
original: tensor(0.0004, device='cuda:0')
new: tensor(3.7944e-05, device='cuda:0')

In some cases, the output image was broken in this case.

Steps to reproduce the problem

1. Create a prompt
2. Generate

What should have happened?

Normal image output

What browsers do you use to access Fooocus?

Other

Where are you running Fooocus?

Locally

What operating system are you using?

Windows 11

Console logs

[Fooocus Model Management] Moving model(s) has taken 2.48 seconds
[Parameters] Adaptive CFG = 7
[Parameters] CLIP Skip = 2
[Parameters] Sharpness = 2
[Parameters] ControlNet Softness = 0.25
[Parameters] ADM Scale = 1.5 : 0.8 : 0.3
[Parameters] Seed = 3825047950370249823
[Parameters] CFG = 4
[Fooocus] Loading control models ...
[Parameters] Sampler = dpmpp_2m_sde_gpu - karras
[Parameters] Steps = 30 - 15
[Fooocus] Initializing ...
[Fooocus] Loading models ...
Refiner unloaded.
[Fooocus] Processing prompts ...
[Fooocus] Preparing Fooocus text #1 ...
[Prompt Expansion] masterpiece, best quality, very aesthetic, 8k, detailed, beautiful color, amazing quality, highres, shiny skin, MGCM, 1girl, solo, (blue eyes, blonde hair, short hair, school uniform, pleated skit, red bow, frills, eriza), walking, street, day, outdoors, bag, skirt, sharp
[Fooocus] Encoding positive #1 ...
[Fooocus Model Management] Moving model(s) has taken 0.29 seconds
[Fooocus] Encoding negative #1 ...
[Parameters] Denoising Strength = 1.0
[Parameters] Initial Latent shape: Image Space (896, 1152)
Preparation time: 2.06 seconds
Using karras scheduler.
[Fooocus] Preparing task 1/1 ...
[Sampler] refiner_swap_method = joint
[Sampler] sigma_min = 0.0291671771556139, sigma_max = 14.614643096923828
Requested to load SDXL
Loading 1 new model
[Fooocus Model Management] Moving model(s) has taken 5.02 seconds
100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00,  1.10it/s]

Additional information

No response