jack-cherish/ai-digital-human

AI-Digital-Human is a project that lets you create a virtual talking character powered by AI. Instead of a generic digital avatar, the idea is to bring a specific person to life, the creator's own demo was a fictionalized "Lu Xun" (a famous Chinese writer) that could hold conversations, speak in a cloned voice, and move its face naturally while talking. It's essentially a toolkit for building your own AI presenter, historical figure, or character companion. The project stitches together several open-source pieces. First, it takes or repairs a face photo so it looks high quality. Then a large language model generates conversational text responses, think of it as the character's "brain." That text gets converted into speech using a voice model you can train on about an hour of someone's recorded audio, so the character sounds like that person. Finally, a tool called SadTalker takes that audio and animates the photo, making the mouth and face move in sync with the words. A simple web interface ties it together so users can chat with the character visually. The target audience is creators, hobbyists, or developers who want to build interactive AI personas without writing everything from scratch. The README points to video tutorials on Bilibili (a Chinese video platform) that walk through setup, including one-click install packages. The README is more of a roadmap than a complete guide right now. The creator admits the project has been delayed and is releasing it in stages across multiple tutorial videos. The original demo used some closed-source tools, so the open-source replacements listed here may produce slightly lower quality results, particularly for the face animation. The full code and setup instructions are promised for later video installments rather than being fully available in the repository itself.