MODEL_DOWNLOAD_GUIDE.md

# Alternative OmniAvatar Model Download Guide

## 🎯 Why You're Getting Only Audio Output

Your app is working correctly but running in **TTS-only mode** because the OmniAvatar-14B models are missing. The app gracefully falls back to audio-only generation when video models aren't available.

## 🚀 Solutions to Enable Video Generation

### Option 1: Use Git to Download Models (If you have Git LFS)

# Create model directories
mkdir pretrained_models\Wan2.1-T2V-14B
mkdir pretrained_models\OmniAvatar-14B  
mkdir pretrained_models\wav2vec2-base-960h

# Clone models (requires Git LFS)
git lfs clone https://huggingface.co/Wan-AI/Wan2.1-T2V-14B pretrained_models/Wan2.1-T2V-14B
git lfs clone https://huggingface.co/OmniAvatar/OmniAvatar-14B pretrained_models/OmniAvatar-14B
git lfs clone https://huggingface.co/facebook/wav2vec2-base-960h pretrained_models/wav2vec2-base-960h

### Option 2: Install Python and Run Setup Script

1. **Install Python** (if not already done):
   - Download from: https://python.org/downloads/
   - Or enable from Microsoft Store
   - Make sure to check "Add to PATH" during installation

2. **Run the setup script**:
   python setup_omniavatar.py

### Option 3: Manual Download from HuggingFace

Visit these URLs and download manually:
- https://huggingface.co/Wan-AI/Wan2.1-T2V-14B
- https://huggingface.co/OmniAvatar/OmniAvatar-14B  
- https://huggingface.co/facebook/wav2vec2-base-960h

Extract to:
- pretrained_models/Wan2.1-T2V-14B/
- pretrained_models/OmniAvatar-14B/
- pretrained_models/wav2vec2-base-960h/

### Option 4: Use Windows Subsystem for Linux (WSL)

If you have WSL installed:
```bash
wsl
cd /mnt/c/path/to/your/project
python setup_omniavatar.py
```

## 📊 Model Requirements

Total download size: ~30.36GB
- Wan2.1-T2V-14B: ~28GB (base text-to-video model)
- OmniAvatar-14B: ~2GB (avatar animation weights)
- wav2vec2-base-960h: ~360MB (audio encoder)

## 🔍 Verify Installation

After downloading, restart your app and check:
- The app should show "full functionality enabled" in logs
- API responses should return video URLs instead of just audio
- Gradio interface should show video output component

## 💡 Current Status

Your setup is working perfectly for TTS! Once the OmniAvatar models are downloaded, you'll get:
✅ Audio-driven avatar videos  
✅ Adaptive body animation
✅ Lip-sync accuracy
✅ 480p video output