ComfyUI
Download and install ComfyUI on Windows.
For real-time updates on verified features and version compatibility, refer to the Juice Application Matrix.
Supported PyTorch Versions
To ensure compatibility with Juice, you must use specific PyTorch builds.
- Supported CUDA Versions:
cu128,cu126. - Unsupported Versions:
13,13.1,13.2.
Installing the Correct PyTorch Build
If you are using a manual or Conda-based installation, run the following command to install the supported cu128 release build:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
Running ComfyUI with Juice
When using a Portable Installation, it is a critical requirement to navigate into the internal ComfyUI folder before executing the run command. This ensures all relative paths for models and custom nodes are resolved correctly.
Portable Installation Execution
cd C:\Path\To\Your\ComfyUI_windows_portable\ComfyUI
juice run python.exe main.py --disable-cuda-malloc
The --disable-cuda-malloc argument is specified for optimal performance with Juice. ComfyUI will still run through Juice without this argument but responses will take longer to generate.
Desktop Application Execution
Run ComfyUI with Juice on Windows using the default application path:
juice run "%USERPROFILE%\AppData\Local\Programs\@comfyorgcomfyui-electron\ComfyUI.exe" --disable-cuda-malloc
Replace %USERPROFILE% with the path to your user directory on Windows, e.g. C:\Windows\joeb, if you prefer.
For 16GB VRAM cards, adding --lowvram, --reserve-vram 2.0 and --use-pytorch-cross-attention to your run command is highly recommended to prevent Out of Memory (OOM) errors during video generation.
Active Paths
Activate the ComfyUI directory to run it through Juice automatically when a Juice session is created:
juice session add-path %USERPROFILE%\AppData\Local\Programs\@comfyorgcomfyui-electron
Replace %USERPROFILE% with the path to your user directory on Windows, e.g. C:\Users\joeb, if you prefer.
Help Improve This Page
Found a mistake? Want to contribute? Edit this page on GitHub or explore more editing options.