Duix (Duix Avatar)
An open-source tool suite that helps content creators and developers generate realistic talking-head videos and interactive digital humans from a short video clip.
Tool overview
Duix is an open-source digital human system by Silicon Intelligence, comprising the offline Duix Avatar video generator and the real-time DUIX interaction framework. With just a ~10-second video, users can clone their appearance and voice locally. It then generates lip-synced videos from text input, supporting 8 languages, all processed on-premises for privacy. The DUIX framework enables voice+vision interaction, runs at 50 FPS on mobile or tablet with only three lines of code, and can be used to create virtual assistants, AI tutors, or digital companions.
Its main advantage is drastically lowering the cost barrier—replacing services that once cost tens of thousands of dollars. A single Docker command and an NVIDIA 3060Ti or better GPU are enough to get started. The open architecture allows developers to integrate their own LLMs, ASR, and TTS for custom experiences.
However, the local pipeline demands a dedicated GPU; ordinary office laptops may struggle. The open-source version focuses on offline video and basic interaction, while commercial cloud services are needed for high-concurrency enterprise scenarios.