Lightricks/LTX-Video-Q8-Kernels

LTXVideo Q8 – q8_kernels

This package implements the operations required to perform inference with the LTXVideo FP8-quantized model.

📚 Table of Contents

• Installation
  • Windows
    • Cloned ComfyUI
    • Portable ComfyUI
  • Linux
  • Troubleshooting
• ComfyUI Integration

🔧 Installation

Note: CUDA Toolkit 12.8 or later is required. The instructions below assume CUDA 12.8 is installed.

🪟 Windows

1. Install the Microsoft Build Tools.
   During installation, select "Desktop development with C++".

2. Make sure the CUDA Toolkit is installed by following the instructions here.

📦 Cloned ComfyUI

1. Run the following commands inside your ComfyUI python environment.

   python.exe -m pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
   
   python.exe -m pip install -U packaging wheel ninja setuptools
   
   python.exe -m pip install --no-build-isolation git+https://github.com/Lightricks/LTX-Video-Q8-Kernels.git
   

📦 Portable ComfyUI

1. Open a Command Prompt in the portable ComfyUI installation folder.

2. Check the version of the embedded Python:

   .\python_embeded\python.exe --version
   

3. Install the matching Python version (e.g., 3.12.10):

   • Download from the official Python website,
     or use pyenv for Windows.

4. Locate your Python installation.
   If installed from official Python website, it is typically located at:
   C:\Users\YOUR_USERNAME\AppData\Local\Programs\Python\Python312

5. Copy the include and libs directories from the full Python installation into the python_embeded folder in your ComfyUI directory.

6. Run the following commands in the Command Prompt:

   .\python_embeded\python.exe -m pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
   
   .\python_embeded\python.exe -m pip install -U packaging wheel ninja setuptools
   
   .\python_embeded\python.exe -m pip install --no-build-isolation git+https://github.com/Lightricks/LTX-Video-Q8-Kernels.git
   

🐧 Linux

1. Make sure the CUDA Toolkit is installed using the instructions here.

   Your CUDA version must match the version used by your installed PyTorch (e.g., CUDA 12.8).

2. Run the following commands in your ComfyUI Python environment:

   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
   pip install packaging wheel ninja setuptools
   pip install --no-build-isolation git+https://github.com/Lightricks/LTX-Video-Q8-Kernels.git
   

🧩 Troubleshooting

• If you encounter issues with the typing-extensions package, run:

  pip install typing-extensions
  

  Then repeat the installation steps.

🎥 ComfyUI Integration

To run inference with ComfyUI:

1. Install the ComfyUI-LTXVideo custom nodes pack.
2. Use the LTXVideo Q8 Patcher node on the loaded model.
3. See the example ComfyUI flow here.