Matrix-Game-2 / README.md
laloadrianmorales's picture
Update README.md
b6cb46d verified
|
raw
history blame
2.76 kB

Matrix-Game-2.0 Hugging Face Space Setup

๐Ÿš€ Quick Setup Guide

1. Create the Space

  • Go to Hugging Face Spaces
  • Click "Create new Space"
  • Choose Gradio as the SDK
  • IMPORTANT: Set Hardware to GPU (A100) - this is crucial!

2. Upload Files

Upload these three files to your Space:

  1. app.py - The main application (use the clean version provided)
  2. requirements.txt - Dependencies list (properly formatted)
  3. README.md - This file (optional)

3. Space Configuration

In your Space settings, ensure:

  • SDK: Gradio
  • Hardware: A100 GPU (required!)
  • Visibility: Public or Private (your choice)

4. Key Improvements Made

โœ… Fixed Issues:

  • Proper requirements.txt formatting (each dependency on new line)
  • Added missing dependencies (opencv, xformers, etc.)
  • Used @spaces.GPU decorator for proper GPU allocation
  • Simplified model downloading and setup
  • Added proper error handling and timeouts
  • Reduced frame limits for HF Spaces constraints

โœ… HF Spaces Optimizations:

  • Pre-download model weights efficiently
  • Use temporary directories properly
  • Clean up resources after generation
  • Progress indicators for user feedback
  • GPU memory management

5. Usage Tips

  • First run: Will take 2-3 minutes to download model (~10GB)
  • Subsequent runs: Much faster (model cached)
  • Recommended: 25-75 frames for reasonable generation time
  • Input images: Works best with landscapes/scenes, max 512px

6. Troubleshooting

If you get "CUDA out of memory":

  • Reduce number of frames (try 25-50)
  • Use smaller input images
  • Restart the Space

If model download fails:

  • Check internet connection in Space logs
  • Verify you're using A100 GPU hardware
  • Try restarting the Space

If generation is slow:

  • This is normal! Video generation is compute-intensive
  • A100 GPU should handle 50 frames in ~2-3 minutes
  • Consider reducing frames for faster results

7. Expected Performance

  • Model download: 2-3 minutes (first time only)
  • 25 frames: ~1 minute generation
  • 50 frames: ~2-3 minutes generation
  • 75+ frames: 4-5+ minutes generation

๐ŸŽฎ What is Matrix-Game-2.0?

Matrix-Game-2.0 is an interactive world foundation model that generates real-time video content based on input images. It uses advanced diffusion techniques to create dynamic, controllable video sequences that respond to user interactions.

Key Features:

  • Real-time video generation (up to 25 FPS)
  • Interactive control through keyboard/mouse inputs
  • Long video generation capability
  • Fine-grained scene control and evolution

Built with โค๏ธ for the Hugging Face community