How to Install FramePack for Video Generation

FramePack is a revolutionary neural network structure for next-frame prediction models in video generation. This guide will walk you through the installation process on both Windows and Linux systems.

Installation Guide for Windows

Quick Installation (Windows)

For Windows users, we provide a convenient one-click package that includes CUDA 12.6 and PyTorch 2.6 pre-configured:

Download One-Click Package (CUDA 12.6 + PyTorch 2.6)

Download the package using the button above
Uncompress the downloaded file to a directory of your choice
Run update.bat to ensure you have the latest version
Run run.bat to start FramePack

Important: Running update.bat is crucial, as it ensures you have the latest version with all bug fixes. Skipping this step may result in using an outdated version with potential issues.

Note: Models will be downloaded automatically during first run. This will require more than 30GB of downloads from HuggingFace. Make sure you have sufficient disk space and a stable internet connection.

Installation Guide for Linux

Linux Installation

For Linux users, we recommend setting up an independent Python 3.10 environment:

1. Install PyTorch with CUDA support:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

2. Install required dependencies:

pip install -r requirements.txt

3. Start the GUI:

python demo_gradio.py

The GUI supports various command-line options including:

--share: Create a public URL
--port: Specify a custom port
--server: Specify server options

Advanced Configuration: Attention Mechanisms

FramePack supports multiple attention mechanisms:

Default PyTorch attention
xformers
flash-attn
sage-attention

By default, FramePack uses PyTorch attention. Advanced users can install alternative attention kernels for potential performance improvements.

Installing sage-attention (Linux)

pip install sageattention==1.0.6

Recommendation: We suggest trying FramePack with the default PyTorch attention first. While alternative attention mechanisms like sage-attention can offer performance improvements, they may slightly influence results.

System Requirements

Minimum requirements for running FramePack:

GPU: NVIDIA GPU with at least 12GB VRAM (16GB+ recommended)
CUDA: CUDA 12.x
RAM: 16GB (32GB recommended)
Storage: 50GB free space for models and cache
Python: Python 3.10

Troubleshooting

If you encounter issues during installation or while running FramePack:

Check that your CUDA drivers are up to date
Ensure you have sufficient disk space for model downloads
Verify that your Python environment is correctly set up (Python 3.10 recommended)
On Windows, try running as administrator if you encounter permission issues
Check the GitHub repository for known issues or submit a new issue

Getting Started

Once installed, you can use FramePack for various video generation tasks:

Text-to-video generation
Image-to-video conversion
Video extension and enhancement
Long-form video creation

Visit our documentation for detailed tutorials and examples.

Next Steps: After installation, check out our Quick Start Guide to create your first AI-generated video with FramePack!

How to install Framepack AI