Installation from ComfyUI manager dosn't work #140
Comments
|
Hi Joe, I have spent all day yesterday trying to install this thing. I'm not a programmer, so I nearly went nuts. Around the same time you posted this, I finally managed to get it to work. I was just about to post an issue, when I saw your message, so I will post it here for anyone else having the same issue. I use Anaconda and Windows OS. Basically, what I managed to piece together is that this doesn't work on Python 3.12 or Python 3.10. You need Python 3.11. Not because of this repo, but because of all the dependencies and the dependencies' dependencies. Using Anaconda, you set up the environment and install Python: conda create --name py311
conda activate py311
conda install python=3.11.9Get CUDA Toolkit v12.1 (this line of code is commented out in this repo, very easy to miss): conda install -c "nvidia/label/cuda-12.1.0" cuda-toolkitGet xformers for CUDA 12.1 (this is adapted from miniconda batch file): python -s -m pip install --force-reinstall xformers --index-url https://download.pytorch.org/whl/cu121Get torch for CUDA 12.1: pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121That was the hardest part. Assuming all this well, the rest basically comes down to running Starting with the main repo: git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
pip install -r requirements.txtComfyUI-Manager: git clone https://github.com/ltdrdata/ComfyUI-Manager custom_nodes
pip install -r custom_nodes/ComfyUI-Manager/requirements.txtAnd lastly, the Comfy3D: git clone https://github.com/MrForExample/ComfyUI-3D-Pack custom_nodes
pip install -r custom_nodes/ComfyUI-3D-Pack/requirements.txt
pip install -r custom_nodes/ComfyUI-3D-Pack/requirements_post.txtNever, under any circumstances, are you to update all your dependencies. That is, apparently, the worst sin Python developer can make. Only update things through ComfyUI-Manager. I hope this helps you and anyone else struggling with this. Good luck! |
|
I guess many of you are having trouble installing Try my all-in-one package: (For Windows) (For Linux) |
|
Unboxed8941 |
|
YanWenKun ImportError: DLL load failed while importing nvdiffrast_plugin: the specified module cannot be found. |
I'm not surprised. I remember that error well, it was the biggest hurdle for me. Ultimately, I got past that point when I deleted my Python 3.12.3 environment and created a new environment with Python 3.11.9. I guess it also depends on other factors like, how much VRAM you have, which OS, do you have the right version of CUDA, does your graphics card support it, is the PyTorch installed, is the PyTorch compiled with the same version of CUDA is, etc... Basically, all the stars have to align for this to work. Good luck! |
Here's an easy (although not elegant) way to fix errors with (Assume you're using Windows 11)
NOTE: Make sure you edited "EDIT_THIS_TO_PATH_TO_YOUR_" This will make sure Ninja (a compiler tool) works. I tested the example CRM workflow on a clean install with a TITAN Xp:
Hope this will have more details covered. Wish you good luck! |
|
YanWenKun |
|
@UnicornJoe42 What's the detail of the subcommand? ```log the long long logs ``` |
|
YanWenKun ''' Error building extension 'nvdiffrast_plugin': [1/6] C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin\nvcc --generate-dependencies-with-compile --dependency-output rasterize.cuda.o.d -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcompiler /EHsc -Xcompiler /wd4068 -Xcompiler /wd4067 -Xcompiler /wd4624 -Xcompiler /wd4190 -Xcompiler /wd4018 -Xcompiler /wd4275 -Xcompiler /wd4267 -Xcompiler /wd4244 -Xcompiler /wd4251 -Xcompiler /wd4819 -Xcompiler /MD -DTORCH_EXTENSION_NAME=nvdiffrast_plugin -DTORCH_API_INCLUDE_EXTENSION_H -IC:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Lib\site-packages\torch\include -IC:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Lib\site-packages\torch\include\torch\csrc\api\include -IC:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Lib\site-packages\torch\include\TH -IC:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include" -IC:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Include -D_GLIBCXX_USE_CXX11_ABI=0 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_89,code=compute_89 -gencode=arch=compute_89,code=sm_89 -std=c++17 -DNVDR_TORCH -lineinfo -c C:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\python_embeded\Lib\site-packages\nvdiffrast\common\rasterize.cu -o rasterize.cuda.o File "C:\3d\ComfyUI_Windows_portable_cu121\ComfyUI_Windows_portable\ComfyUI\execution.py", line 151, in recursive_execute ''' |
|
@UnicornJoe42 |
|
@UnicornJoe42 |
Thank you! :) |
|
Hello! I'm using your repo: https://github.com/YanWenKun/ComfyUI-Windows-Portable Error occurred when executing [Comfy3D] Load Convolutional Reconstruction Model: Ninja is required to load C++ extensions File "D:\ComfyUI_Windows_portable\ComfyUI\execution.py", line 151, in recursive_execute error log[Comfy3D] [Load_CRM_MVDiffusion_Model] loaded model ckpt from D:\ComfyUI_Windows_portable\ComfyUI\custom_nodes\ComfyUI-3D-Pack\checkpoints/crm\ccm-diffusion.pth !!! Exception during processing !!! Prompt executed in 41.87 seconds
I'm not a programmer, but it seems as though the script can not find python: the "a", as the beginning invalidates the path: '"D:\a\ComfyUI-Windows-Portable\ComfyUI-Windows-Portable\python_embeded\python.exe" "D:\ComfyUI_Windows_portable\python_embeded\Scripts\ninja.exe" --version': The system cannot find the file specified. Any idea how to fix? |
|
@rethink-studios I guess something went wrong with PIP's Ninja, so I updated the package and replaced If you don't want to download the whole package again, just download the Ninja binary, extract it to |
How do i use the .sh files? As readme states "Run |
https://github.com/YanWenKun/ComfyUI-Windows-Portable/releases |
Thanks mate :) So, a few bumps on that road once extracted and launched. As well as this error:
Tested another workflow from 3d pack, and now got this.
|
Just ignore this. It's not related to 3D-Pack. For workflows: Try TripoSR first, it's more likely to work. For zero123plus: Click "Manager" on the right side panel, then click "Update All" and then click "RESTART" when done updating. |
Updated all now, and neither of these will open at all. CRM MVDiffusion_to_InstantMesh.json These ones does load [FlexiCubes] Multi-View Depth plus Optional Normal maps to 3D Mesh.json Zero can successfully generate 3d models, but any other diffuser pipeline scheduler will crash, and not work again until you load the workflow json anew But wonder3D errors out on generation I tested "model impactframes__if_3d_crm_playground__comfyworkflows.json" from a youtuber, it also works, albeit very low quality 3d models and low textures. |
The 3D-Pack will try to download the models when needed. But errors may happen. Just download the model from HF and put it in: The same goes with other "Model file not found" error. Just google the model/repo name and put them under |
thanks :) But what about the other workflows that won't load in at all? And for wonder3d, now model complaints are gone, but i've faced this error before, but never managed to fix it, what does it mean by "keys"?
|
Check HF, download/clone the whole repo into
Need console output to know what happened. |
|
I have a similar problem but wondering if it's related to Stability Matrix and how it sets up my different packages/platforms. The only error that seems to be happening is this: I've tried both methods of install just to be sure. First method, I'm unsure because I don't think I had to since I have VS Studio 2022 and all of the C++ extentions and such for building CUDA things (pretty much everything windows C++ also), and the link just takes me to where you download VS 2022. I did everything for the miniconda and made the environment, but pip install -r requirements_post.txt fails to install giving the error: Will try the alternatives listed here at some point. It's just dumb since it's probably one or two small things that aren't downloading properly... sigh
That's cool but I have no need for all of these things and like to keep only what I need. I prefer sniper rifles to shotguns. If it was for a lot more than just the 3D-pack stuff maybe. In fact I probably should just figure out which single 3D workflow suits my needs and system and just manually get the custom nodes for that I suppose. |
The text was updated successfully, but these errors were encountered: