PersonaLive

Why we included this project

PersonaLive turns a single portrait photo into an expressive talking-head video you can drive in real time. It's a streamable diffusion framework that keeps the reference face's identity and lip sync intact while you animate the avatar from a webcam and broadcast continuously, instead of rendering short clips one at a time. The repo ships inference code, pretrained weights, training code, and a WebUI, and its streaming strategy can generate long, unbounded videos on a single 12GB GPU. ComfyUI support lets it slot into existing generation pipelines. That makes it a practical option for streamers, VTuber-style creators, and developers prototyping live avatar apps, with the caveat that the project is released for academic research only.