Wan 2.2 Google Colab Template

A ready to use Google Colab notebook template for running Wan 2.2, an advanced image to video AI model. This template simplifies the setup process and provides an optimized workflow for generating videos from images on Colab's free GPU resources.

Features

• One Click Setup: Automated installation of all dependencies and model downloads
• Optimized for Colab: Pre-configured settings for Google Colab (also works on other notebook platforms)
• Memory Efficient: Includes flags for model offloading and CPU-based text encoding to manage VRAM
• User Friendly Interface: Simple upload → configure → generate workflow
• Multiple Resolutions: Support for 1280×720, 832×480, and 1024×1024 outputs

Quick Start

Option 1: Using the Notebook (Recommended)

1. Open in Google Colab:

   • Upload wan2_2.ipynb to Google Colab, or
   • Use the "Open in Colab" button if you've set one up

2. Select GPU Runtime:

   • Go to Runtime → Change runtime type
   • Select GPU (preferably A100 if available)
   • Click Save

3. Run the Cells:

   • Execute each cell in order (or use Runtime → Run all)
   • The notebook will:
     • Clone the Wan 2.2 repository
     • Install all required dependencies
     • Download the model (≈60GB, takes 5ish minutes)
     • Set up the generation pipeline

4. Generate Your Video:

   • Upload an input image when prompted
   • Configure your prompt and settings
   • Run the generation cell and wait for your video!

Option 2: Using the Python Script

If you prefer working with Python scripts directly:

# The wan2_2.py file contains the same workflow
# You can run it in any Python notebook environment

Requirements

Recommended Hardware (Google Colab)

• GPU: NVIDIA A100 (40GB) or T4 (16GB)
• RAM: 12GB+ system RAM
• Storage: 70GB+ free space for model and temporary files

Software Dependencies

All dependencies are automatically installed by the notebook:

• Python 3.8+
• PyTorch 2.4.0+
• Transformers, Diffusers, Accelerate
• Flash Attention
• Various CV and ML libraries

Usage Guide

Step 1: Upload Your Image

The notebook will prompt you to upload an input image. Supported formats:

• JPEG/JPG
• PNG
• BMP

Step 2: Configure Settings

Prompting

Write a detailed text description of the motion/action you want to see in your video:

Example: "A woman smiling and waving at the camera, 
her hair gently moving in the breeze"

Model Configuration

• Task: Choose between i2v-A14B (14B parameters, higher quality) or i2v-1.3B (faster)
• Resolution: Select from 1280*720, 832*480, or 1024*1024
• Checkpoint Directory: Default is ./Wan2.2-I2V-A14B

Optimization Flags

Enable these to manage memory usage:

• Offload Model: Offloads parts of the model to CPU to save VRAM
• Use T5 CPU: Runs text encoder on CPU (recommended for free tier)
• Convert Model Dtype: Converts model to lower precision for memory efficiency

Step 3: Generate Video

Run the generation cell and wait. Processing typically takes:

• A100 GPU: 5-10 minutes per video
• T4 GPU: 15-25 minutes per video

Step 4: View Your Video

The notebook automatically displays the generated video inline when complete.

Repository Structure

wan2.2-google-colab/
├── wan2_2.ipynb          # Main Colab notebook
├── wan2_2.py             # Python script version
└── README.md             # This file

Customization

Changing the Model

To use a different Wan 2.2 checkpoint:

1. Update the checkpoint_dir variable in Step 2
2. Use the appropriate task identifier (i2v-A14B or i2v-1.3B)

Advanced Parameters

You can modify the generation command in Step 3 to add more flags. Refer to the official Wan 2.2 repository for all available options.

Troubleshooting

Out of Memory Errors

If you encounter CUDA out of memory errors:

1. Enable all optimization flags (offload, T5 CPU, convert dtype)
2. Try a lower resolution (832×480 instead of 1280×720)
3. Use the smaller 1.3B model instead of 14B
4. Restart runtime and clear cache before running

Model Download Fails

• Check your internet connection
• The model is ≈60GB, ensure enough storage
• Try rerunning the download cell

Video Quality Issues

• Use more detailed prompts
• Ensure input image is high quality
• Try the larger A14B model if using 1.3B
• Experiment with different resolutions

Credits

• Wan 2.2 Model: Wan-Video/Wan2.2

License

This template follows the same license as the Wan 2.2 project. Please refer to the original repository for licensing information.

Contributing

Contributions are welcome! If you have improvements or fixes:

1. Fork this repository
2. Make your changes
3. Submit a pull request

Additional Resources

• Wan 2.2 Official Repository
• Hugging Face Model
• Google Colab Documentation

Support

For issues specific to this template, please open an issue in this repository.

For questions about the Wan 2.2 model itself, refer to the official repository.

---

Note: This is an unofficial template designed to make Wan 2.2 more accessible on Google Colab and other notebook platforms. Always ensure you have appropriate compute resources and comply with the platform's usage policies.