Easily install ComfyUI (in a python venv) on Linux. Tested on Arch + AMD GPU.
Note: This is not the official ComfyUI icon.
Make sure to install for your GPU Vendor (AMD/Nvidia):
git clone https://github.com/HAMM3REXTREME/ComfyUI-Installer
cd ComfyUI-Installer
./install-amd.sh or ./install-nvidia.sh # Optionally add --menu-entry
./launch.sh # Launch
-
To install ComfyUI using this script, clone this repo and cd into it:
git clone https://github.com/HAMM3REXTREME/ComfyUI-Installer && cd ComfyUI-Installer
-
After that is completed, install for either AMD or Nvidia.
To install for AMD:./install-amd.sh
To install for Nvidia:./install-nvidia.sh
Tip: You can optionally add--menu-entry
after your GPU vendor in order to make a desktop entry. -
Almost done! You can now get your models and put them in their proper directories:
Put your SD checkpoints (the huge ckpt/safetensors files) in:ComfyUI/models/checkpoints
Put your VAE in:ComfyUI/models/vae
Once that is done, launch ComfyUI using: ./launch.sh
Simply cd into the ComfyUI folder and run git pull:
cd ComfyUI && git pull
Make sure to check out the actual ComfyUI repo for the most up to date information.
-
Run
source venv/bin/activate
to activate the Python virtual environment.
(The installer creates a python venv named 'venv' in the base folder by default) -
Upgrade torch, use the command for your GPU vendor (similar to the installer script):
For AMD:pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0
For Nvidia:pip install --upgrade torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121
Some version numbers might be out of date here. -
Upgrade ComfyUI dependencies by running this command (inside the ComfyUI folder):
pip install --upgrade -r requirements.txt
After this you should have everything updated and can proceed to running ComfyUI.
If you get the "Torch not compiled with CUDA enabled" error, uninstall torch with:
pip uninstall torch
And install it again with the command above.
For AMD cards not officially supported by ROCm, Try running it with this command if you have issues:
For 6700, 6600 and maybe other RDNA2 or older: HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py
This is the done by default in the launch script. Feel free to remove/edit it if you want.
For AMD 7600 and maybe other RDNA3 cards: HSA_OVERRIDE_GFX_VERSION=11.0.0 python main.py