[Bug]: Issue Running V-Pred Model on A1111 Dev Branch: Black Output After Fixing NaNs Error #17067
Description
Checklist
- The issue exists after disabling all extensions
- The issue exists on a clean installation of webui
- The issue is caused by an extension, but I believe it is caused by a bug in the webui
- The issue exists in the current version of the webui
- The issue has not been reported before recently
- The issue has been reported before but has not been fixed yet
What happened?
I'm trying to run a V-pred model:
https://files.catbox.moe/vciz3c.pdf
I followed the instructions to use the dev branch of Automatic1111 to make it work. However, after a clean install, I encountered this error:
NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion, or use the --no-half command line argument to fix this. Use --disable-nan-check to disable this check.
After I added the --no-half and --disable-nan-check arguments, the error disappeared — but the model only generates black images.
Is there any additional step required to get this model working correctly?
My Specs :
NVIDIA GeForce RTX 3090 (24GB VRAM), 64GB RAM, Intel Core i5-9600K
I’ve also uploaded my full system information file generated by the WebUI below.
Steps to reproduce the problem
I performed a clean install of the WebUI using the dev branch of Automatic1111, following the instructions provided in this document:
https://files.catbox.moe/vciz3c.pdf
Then, I tried generating an image using the V-Pred model mentioned in the guide.
What should have happened?
its generate an image of Model Sdxl V-Pred
What browsers do you use to access the UI ?
Google Chrome
Sysinfo
Console logs
Here is the full Console :
Running on local URL: http://127.0.0.1:7861
To create a public link, set `share=True` in `launch()`.
Creating model from config: D:\stable-diffusion-webui-dev\configs\sd_xl_v.yaml
D:\stable-diffusion-webui-dev\venv\lib\site-packages\huggingface_hub\file_download.py:943: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
warnings.warn(
Startup time: 14.9s (prepare environment: 3.0s, import torch: 6.0s, import gradio: 1.3s, setup paths: 1.7s, initialize shared: 0.3s, other imports: 0.4s, load scripts: 0.8s, create ui: 0.7s, gradio launch: 0.6s).
Applying attention optimization: xformers... done.
Model loaded in 12.1s (load weights from disk: 1.0s, create model: 0.5s, apply weights to model: 9.8s, apply half(): 0.1s, apply dtype to VAE: 0.2s, move model to device: 0.1s, calculate empty prompt: 0.2s).
40%|████████████████████████████████▊ | 12/30 [00:03<00:05, 3.48it/s]D:\stable-diffusion-webui-dev\modules\sd_samplers_common.py:68: RuntimeWarning: invalid value encountered in cast54it/s]
x_sample = x_sample.astype(np.uint8)
100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:08<00:00, 3.44it/s]
*** Error completing request███████████████████████████████████████████████████████████| 30/30 [00:08<00:00, 3.46it/s]
*** Arguments: ('task(0zf4yw7wrdpmybr)', <gradio.routes.Request object at 0x000001805EA70FA0>, 'masterpiece, best quality, newest, absurdres, highres, nsfw, anime screencap, ayase seiko, dandadan, 1girl, mature female, red eyes, glasses, beehive hairdo, long hair, white hair, grey hair, ponytail, large breasts, cleavage, thighs, hoop earrings, baseball bat, cigarette, mature female,', 'sfw, worst quality, old, early, low quality, lowres, signature, username, logo, bad hands, mutated hands, mammal, anthro, furry, ambiguous form, feral, semi-anthro, 3d, realistic,', [], 1, 1, 7, 1216, 864, False, 0.7, 2, 'Latent', 0, 0, 0, 'Use same checkpoint', 'Use same sampler', 'Use same scheduler', '', '', ['Downcast alphas_cumprod: True', 'Clip skip: 2'], 0, 30, 'Euler a', 'Automatic', False, '', 0.8, 696349519, False, -1, 0, 0, 0, False, False, 'positive', 'comma', 0, False, False, 'start', '', 1, '', [], 0, '', [], 0, '', [], True, False, False, False, False, False, False, 0, False, True) {}
Traceback (most recent call last):
File "D:\stable-diffusion-webui-dev\modules\call_queue.py", line 74, in f
res = list(func(*args, **kwargs))
File "D:\stable-diffusion-webui-dev\modules\call_queue.py", line 53, in f
res = func(*args, **kwargs)
File "D:\stable-diffusion-webui-dev\modules\call_queue.py", line 37, in f
res = func(*args, **kwargs)
File "D:\stable-diffusion-webui-dev\modules\txt2img.py", line 109, in txt2img
processed = processing.process_images(p)
File "D:\stable-diffusion-webui-dev\modules\processing.py", line 847, in process_images
res = process_images_inner(p)
File "D:\stable-diffusion-webui-dev\modules\processing.py", line 998, in process_images_inner
devices.test_for_nans(samples_ddim, "unet")
File "D:\stable-diffusion-webui-dev\modules\devices.py", line 265, in test_for_nans
raise NansException(message)
modules.devices.NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.Additional information
No response