Your digital clone, running offline on your own GPU

What it does

Duix.Avatar is a desktop application paired with Dockerized backend services that generates AI avatar videos entirely offline. You feed it video and audio of a real person, and it builds a model that can be driven by text or voice to produce new lip-synced footage. The stack handles voice cloning, speech recognition, and facial motion synthesis locally, targeting Windows and Ubuntu desktops with NVIDIA GPUs.

The interesting bit

The project comes from Duix.com, a company that spent seven years commercializing this technology before releasing it as open source—an unusual reverse path from paid tool to free toolkit. Its fully-offline architecture is pitched as a privacy play, keeping biometric data on-device rather than in the cloud, though that freedom demands serious local hardware.

Key highlights

• Runs entirely offline via local Docker containers; no internet required after setup.
• Supports eight languages for scripts: English, Japanese, Korean, Chinese, French, German, Arabic, and Spanish.
• Exposes local HTTP APIs for model training, audio synthesis, and video generation.
• Ships with a desktop client installer for Windows and an AppImage for Ubuntu.
• Offers a lite Docker deployment that drops some services for smaller footprints.

Caveats

• Hardware appetite is steep: the README recommends an RTX 4070, 32 GB of RAM, and over 100 GB of free disk space just for images.
• API documentation in the README is truncated and incomplete; you’ll need to read the source files to understand the contracts.
• Windows support is narrowly gated to build 19042.1526 or newer, and an NVIDIA GPU is effectively mandatory.

Verdict

Content creators, educators, or developers who need private, local avatar generation and have a modern NVIDIA rig should look here. If you’re on macOS, running older hardware, or looking for a lightweight web API, this is not your tool.