Unable to launch ComfyUI the second time

[Kazhamania]

Hi everyone,
I'm a new user of ComfyUI and I have a problem to fix !
Each time I do a fresh install of ComfyUi, I can use the program as well. When I close the program and try to launch it a second time, I can't.
I have this error :

E:\ComfyUI\python_embeded\Lib\site-packages\torch\cuda\__init__.py:147: UserWarning: Failed to initialize NumPy: _ARRAY_API not found (Triggered internally at ..\torch\csrc\utils\tensor_numpy.cpp:84.) torch.tensor([1.0], dtype=torch.bfloat16, device=device)

Any help ?

[LukeG89]

Can you check your NumPy version?
Open terminal in your main folder (where run_nvidia_gpu.bat is located) and run this command:

python_embeded\python.exe -m pip show numpy

If version is higher than 2, you need to downgrade it; there are issues with that. Use this command:

python_embeded\python.exe -m pip install numpy<2 --force-reinstall

[Kazhamania]

Ok, I will.

I checked :
Version: 2.1.3

With your second commande I have no result :
E:\ComfyUI>python_embeded\python.exe -m pip install numpy<2 --force-reinstall Le fichier spécifié est introuvable. (translation : The specified file could not be found)

[LukeG89]

Try uninstalling first:

python_embeded\python.exe -m pip uninstall numpy

And then installing again:

python_embeded\python.exe -m pip install numpy<2

[Kazhamania]

Thanks for reply, I do :

E:\ComfyUI>python_embeded\python.exe -m pip uninstall numpy
Found existing installation: numpy 2.1.3
Uninstalling numpy-2.1.3:
  Would remove:
    e:\comfyui\python_embeded\lib\site-packages\numpy-2.1.3-cp312-cp312-win_amd64.whl
    e:\comfyui\python_embeded\lib\site-packages\numpy-2.1.3.dist-info\*
    e:\comfyui\python_embeded\lib\site-packages\numpy.libs\libscipy_openblas64_-c16e4918366c6bc1f1cd71e28ca36fc0.dll
    e:\comfyui\python_embeded\lib\site-packages\numpy.libs\msvcp140-d64049c6e3865410a7dda6a7e9f0c575.dll
    e:\comfyui\python_embeded\lib\site-packages\numpy\*
    e:\comfyui\python_embeded\scripts\f2py.exe
    e:\comfyui\python_embeded\scripts\numpy-config.exe
Proceed (Y/n)? y
  Successfully uninstalled numpy-2.1.3

E:\ComfyUI>python_embeded\python.exe -m pip install numpy<2
Le fichier spécifié est introuvable.

(translation : The specified file could not be found)

Successful uninstalled, but unable to re-install. This is strange.

[LukeG89]

Try installing this version (it's the same as I have)

python_embeded\python.exe -m pip install numpy==1.26.4

[ltdrdata]

Le fichier spécifié est introuvable.
Translation:

I haven't tested it on Windows, but you might need to escape the <.

python_embeded\python.exe -m pip install numpy\<2

[Kazhamania]

Hi, thanks for all reply,
It's seems that I have installed an older version of numpy now
Version: 1.26.4

But, when I launch the run_nvidia_gpu, I have no result :

E:\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Total VRAM 8192 MB, total RAM 32721 MB
pytorch version: 2.3.1+cu121
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2070 SUPER : cudaMallocAsync
Using pytorch cross attention
[Prompt Server] web root: E:\ComfyUI\ComfyUI\web

E:\ComfyUI>pause

[Kazhamania]

Sure,

I have done this update and check my numpy version

Version: 1.26.4

But I still have the same "no result" when launching run_nvidia_gpu :

E:\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Total VRAM 8192 MB, total RAM 32721 MB
pytorch version: 2.5.1+cu124
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2070 SUPER : cudaMallocAsync
Using pytorch cross attention
[Prompt Server] web root: E:\ComfyUI\ComfyUI\web

E:\ComfyUI>pause

[LukeG89]

That's really strange... @ltdrdata What do you think?

[Kazhamania]

Maybe we can start the process with a fresh install step by step ?

[LukeG89]

You can download the latest ComfyUI portable from here and see if you encouter any problems when you run it.
Then you can move all your files (models, input, output, workflows, etc.) to the new installation.

[ltdrdata]

I also think it would be a good idea to try a clean install.

[Kazhamania]

Ok,

I have done a fresh install of ComfyUI.
I have not launch the software.

The numpy version is > 2, so I have uninstalled this version and do a fresh install of numpy 1.26.4.

I launch ComfyUI the first with run_nvidia_gpu = work

I launch ComfyUI a second time = work

I install ComfyUI Manager on my custom_nodes folder.

I launch ComfyUI a third time = work

I edit my extra_model_paths.yalms file to link my Stable Diffusion ressources.

I launch ComfyUI a fourth time = NOT WORK

E:\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Total VRAM 8192 MB, total RAM 32721 MB
pytorch version: 2.5.1+cu124
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2070 SUPER : cudaMallocAsync
Using pytorch cross attention
[Prompt Server] web root: E:\ComfyUI\ComfyUI\web
Adding extra search path checkpoints E:\StableDiffusion\stable-diffusion-webui\models/Stable-diffusion
Adding extra search path configs E:\StableDiffusion\stable-diffusion-webui\models/Stable-diffusion
Adding extra search path vae E:\StableDiffusion\stable-diffusion-webui\models/VAE
Adding extra search path loras E:\StableDiffusion\stable-diffusion-webui\models/Lora
Adding extra search path loras E:\StableDiffusion\stable-diffusion-webui\models/LyCORIS
Adding extra search path upscale_models E:\StableDiffusion\stable-diffusion-webui\models/ESRGAN
Adding extra search path upscale_models E:\StableDiffusion\stable-diffusion-webui\models/RealESRGAN
Adding extra search path upscale_models E:\StableDiffusion\stable-diffusion-webui\models/SwinIR
Adding extra search path embeddings E:\StableDiffusion\stable-diffusion-webui\embeddings
Adding extra search path hypernetworks E:\StableDiffusion\stable-diffusion-webui\models/hypernetworks
Adding extra search path controlnet E:\StableDiffusion\stable-diffusion-webui\models/ControlNet
E:\ComfyUI\python_embeded\Lib\site-packages\kornia\feature\lightglue.py:44: FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead.
  @torch.cuda.amp.custom_fwd(cast_inputs=torch.float32)

[Kazhamania]

Does your statement mean that it worked before editing the extra_model_paths.yaml file, but it stopped working after the edit and cannot be restored even after deleting the file?

Yes, that's it !

[Kazhamania]

Any ideas ? ^^

[ltdrdata]

Based on your description, it seems that the issue occurs randomly.

It might be a good idea to perform a clean install and run repeated tests without making any changes.

[Kazhamania]

Based on your description, it seems that the issue occurs randomly.

It might be a good idea to perform a clean install and run repeated tests without making any changes.

Ok, I have made a fresh install and instead of editing my extra_model_paths.yaml , I have directly copied the sources that I use for A111 (models, VAE etc..) inside the ComfyUI folder and .... problem is here again !

E:\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Total VRAM 8192 MB, total RAM 32721 MB
pytorch version: 2.5.1+cu124
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2070 SUPER : cudaMallocAsync
Using pytorch cross attention
[Prompt Server] web root: E:\ComfyUI\ComfyUI\web
E:\ComfyUI\python_embeded\Lib\site-packages\kornia\feature\lightglue.py:44: FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead.
  @torch.cuda.amp.custom_fwd(cast_inputs=torch.float32)

E:\ComfyUI>pause

[LukeG89]

At this point my last guess is... Could it be a storage problem? Your E:\ drive has corrupted memory or something?

Have you tried a different drive?