A high-performance digital human video generation application with GPU acceleration support for NVIDIA RTX series graphics cards.

This project transforms a CPU-only digital human application into a dual-environment system supporting both CPU and GPU processing, achieving 40% performance improvement on NVIDIA RTX hardware.

• 🚀 GPU Acceleration: CUDA 11.8 + PyTorch 2.0.1 optimization
• 🔄 Dual Environment: CPU (stable) + GPU (performance) options
• 🌐 Web Interface: Gradio-based UI with real-time processing
• 📱 CLI Support: Command-line tools for batch processing
• 🛡️ Error Recovery: Graceful fallbacks and comprehensive error handling

• OS: Linux Ubuntu 18.04+
• Python: 3.8+
• RAM: 8GB+
• Storage: 10GB free space

• GPU: NVIDIA RTX 20/30/40 series
• VRAM: 6GB+ recommended
• CUDA: 11.8+ compatible drivers
• Driver: NVIDIA 450.80.02+

git clone https://github.com/agilealpha1/AiVideo.git
cd AiVideo

# Create CPU environment
python -m venv venv
source venv/bin/activate

# Install CPU dependencies
pip install -r requirements_updated.txt

# Run CPU web interface
python app.py
# Access at: http://localhost:7860

# Create GPU environment
python -m venv venv_gpu
source venv_gpu/bin/activate

# Install GPU-optimized PyTorch
pip install torch==2.0.1+cu118 torchaudio==2.0.2+cu118 torchvision==0.15.2+cu118 --index-url https://download.pytorch.org/whl/cu118

# Install other dependencies
pip install -r requirements_gpu_fixed.txt

# Additional required packages
pip install einops typeguard==2.13.3

# Run GPU web interface
python app_gpu.py
# Access at: http://localhost:7861

1. Upload Files: Audio (.wav) + Video (.mp4)
2. Select Code: Processing identifier (default: 1004)
3. Choose Environment:
   • CPU: http://localhost:7860 (stable)
   • GPU: http://localhost:7861 (faster)
4. Process: Click "Process" button
5. Download: Get generated video/audio files

# CPU processing
source venv/bin/activate
python run.py --audio_path audio.wav --video_path video.mp4

# GPU processing
source venv_gpu/bin/activate
python run_gpu.py --audio_path audio.wav --video_path video.mp4 --gpu

# Force CPU mode in GPU environment
python run_gpu.py --audio_path audio.wav --video_path video.mp4 --cpu

HeyGem-Linux-Python-Hack/
├── 📁 CPU Environment
│   ├── app.py                    # CPU web interface
│   ├── run.py                   # CPU command line
│   └── requirements_updated.txt # CPU dependencies
├── 📁 GPU Environment  
│   ├── app_gpu.py              # GPU web interface
│   ├── run_gpu.py              # GPU command line
│   └── requirements_gpu_fixed.txt # GPU dependencies
├── 📁 Core Modules
│   ├── service/                # Core processing logic
│   ├── face_lib/              # Face detection/processing
│   ├── landmark2face_wy/      # Neural network models
│   └── y_utils/               # Utility functions
├── 📁 Configuration
│   ├── config/                # Application settings
│   └── example/               # Sample input files
└── 📁 Documentation
    ├── README.md              # This file
    └── .gitignore            # Git exclusions

RuntimeError: Unexpected error from cudaGetDeviceCount()

Solution: Use CPU environment or update NVIDIA drivers

OSError: Cannot find empty port in range: 7860-7860

Solution: Check for running processes:

ps aux | grep python
kill <process_id>  # If needed

ModuleNotFoundError: No module named 'einops'

Solution: Install missing packages:

pip install einops typeguard==2.13.3

_queue.Empty: timeout

Status: Non-critical - processing continues, files still generated

# Clear GPU cache
python -c "import torch; torch.cuda.empty_cache()"

nvidia-smi

# Test CUDA availability
python -c "import torch; print(f'CUDA: {torch.cuda.is_available()}')"

# Test web interface
curl http://localhost:7860  # CPU version
curl http://localhost:7861  # GPU version

# Compare processing times
time python run.py --audio_path example/audio.wav --video_path example/video.mp4
time python run_gpu.py --audio_path example/audio.wav --video_path example/video.mp4 --gpu

• torch>=2.0.1 - Deep learning framework
• gradio>=4.44.1 - Web interface
• opencv-python>=4.7.0 - Computer vision
• numpy>=1.21.6,<1.23.0 - Numerical computing
• scipy>=1.7.1,<1.8.0 - Scientific computing

• torch==2.0.1+cu118 - CUDA-enabled PyTorch
• onnxruntime-gpu==1.19.2 - GPU inference runtime
• einops==0.8.1 - Tensor operations
• typeguard==2.13.3 - Type checking

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push branch (git push origin feature/amazing-feature)
5. Open Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

• Original HeyGem Digital Human team for the base application
• @Holasyb918/HeyGem-Linux-Python-Hack - Docker-free Linux Python implementation
• @duixcom/Duix.Heygem - Enhanced HeyGem implementation
• NVIDIA for CUDA toolkit and GPU optimization guides
• PyTorch team for GPU acceleration framework
• Gradio team for the excellent web interface framework

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: See /docs folder for detailed guides

⭐ Star this repository if it helped you optimize your digital human processing!