# Wan2.2 Smoke Training Package

Generated: 2026-05-13 13:58:31

## Contents

- `clips/` - 160 training video clips
- `first_frames/` - First frame for each clip
- `diffsynth_metadata.jsonl` - Training metadata
- `train.py` - Training script
- `launch.sh` - Launch script
- `requirements.txt` - Python dependencies

## Usage on Cloud GPU Provider

1. Upload this package to your GPU instance
2. Download Wan2.2 model:
   ```bash
   huggingface-cli download Wan-AI/Wan2.2-TI2V-5B --local-dir /models/Wan2.2-TI2V-5B
   ```
3. Run training:
   ```bash
   bash launch.sh
   ```

## Expected Resource Usage

- GPU: RTX 4090 or better (24GB+ VRAM)
- Training time: ~2-3 hours for smoke test
- Disk space: ~5GB for package + ~15GB for model

## Provider Recommendations

- **RunPod**: RTX 4090 @ $0.70/hr
- **Vast.ai**: RTX 4090 @ $0.40-0.60/hr
- **Lambda Labs**: A100 @ $1.50/hr

Estimated cost: $2-5 for full smoke test