Duix-Avatar

Duix.Avatar is an open-source desktop tool for creating AI-generated videos of a realistic digital copy of yourself. You provide a short video clip of your face and a voice sample, and the software builds a virtual model that can be made to say whatever you write or record. The result is a video of your likeness speaking the content you provided.

The tool runs entirely on your local computer with no internet connection required. All processing happens on your machine, so no footage, voice samples, or generated videos are sent to any server. It runs on Windows 10 (version 19042 or later) and Ubuntu 22.04.

Setup is Docker-based. You download several Docker images totaling roughly 70 gigabytes and run them with a single command. The system also requires an NVIDIA graphics card, at least 32 gigabytes of RAM, and substantial disk space (around 130 gigabytes combined for the Windows version).

Once running, you interact through a desktop application where you upload your appearance and voice samples, then type or speak the text you want the avatar to deliver. The avatar lip-syncs to the audio and reflects natural speech patterns. Scripts can be written in eight languages: English, Japanese, Korean, Chinese, French, German, Arabic, and Spanish.

The company behind the project, Duix.com, says it previously sold this technology commercially and that more than 10,000 businesses have used it. The open-source release makes the same underlying system available at no cost. A real-time interactive mode is also available through a separate open API for developers who want to build conversational avatar applications.