Duix.Avatar AI Avatar Toolkit

Duix.Avatar is an open-source AI avatar toolkit designed for offline video generation and digital human cloning. It allows users to clone their appearance and voice using a short video (approximately 10 seconds) and then drive the digital avatar using text or uploaded audio to generate talking-head videos.

Core Capabilities

• Appearance & Voice Cloning: High-precision capture of facial features and vocal characteristics.
• Text & Audio Driving: Uses NLP to convert text to speech or accepts audio input to synchronize lip movements and expressions.
• Efficient Video Synthesis: Synchronizes video frames with audio for natural lip-syncing.
• Multi-language Support: Supports English, Japanese, Korean, Chinese, French, German, Arabic, and Spanish.
• Offline Operation: Designed for local deployment to ensure privacy and data security.

When adding a new model, the source video must meet these criteria:

• The video must contain audio.
• The audio must feature a person speaking.

This is required because the program uses the audio from the video to perform voice cloning.

Duix Avatar offers two distinct ways to use the technology:

For enterprise-level SLA and high-definition effects, use the Digital Human/Clone Voice API Service via https://docs.duix.com.

Prepare a voice model for cloning by following these steps:

1. Prepare Data: Separate your source video into a silent video and its corresponding audio track.
2. Organize Files: Place the audio files in the designated data directory. By default, this is D:\duix_avatar_data\voice\data (aligned with the guiji2025/fish-speech-ziming service), but this can be modified in your docker-compose configuration.
3. Execute Training: Call the Model Training API (refer to src/main/service/model.js for implementation details).
4. Capture Results: Record the API response. You will need the returned asr_format_audio_url and reference_audio_text to perform subsequent Audio Synthesis tasks.

If you are using an NVIDIA 50-series GPU (or 30/40-series users with CUDA 12.8), use the specialized deployment configuration which utilizes a preview version of PyTorch.

Run the following command from the /deploy directory:

cd /deploy
docker-compose -f docker-compose-5090.yml up -d

When reporting an issue, provide detailed reproduction steps and error logs:

• Client Logs: Follow the method shown in the project documentation (refer to the visual guide in the repository).
• Server Logs: Locate the specific error in the logs, or access the three Docker services and use the "Copy" function to retrieve the log output.

Before deploying Duix.Avatar locally, ensure the following dependencies are installed on your system:

1. Node.js: Version 18 or higher.
2. Docker Images: You must pull the following required images:
   • guiji2025/fun-asr
   • guiji2025/fish-speech-ziming
   • guiji2025/duix.avatar

docker pull guiji2025/fun-asr
docker pull guiji2025/fish-speech-ziming
docker pull guiji2025/duix.avatar

For users who want to experience Duix.Avatar without complex local deployment, digital human cloning agents and plugins are available on the Coze platform. You can directly use the following:

• Cloning Agent: Link to Coze Store
• Cloning Plugin: Link to Coze Store

The src/main/index.js file serves as the entrypoint for the Electron main process. Upon startup, the application performs several critical initialization steps:

1. Database Initialization: Calls initDB() to set up the local data storage.
2. Interval Initialization: Calls initInterval() to manage background tasks or periodic operations.
3. Service Registration: Executes registerHandler() to register core application services.
4. Web Interface Registration: Executes registerWebHandles({ mainWindow }) to expose specific client interfaces to the renderer process.

Developers extending the application should ensure that any new core services are registered within the app.whenReady() lifecycle to ensure the environment is fully prepared.