daniloespeleta/PersonaLive ? reverse-engineered prompt

Build me a simple app that animates a portrait photo in real time from a driving video or webcam feed, so I can upload one reference image and see it move, blink, and talk live in the browser.

Please wire up the existing inference code so the offline mode can generate a video from a reference image and driving video, and the online mode can run as a local web page with a clean upload flow and a start button. If it helps, make the setup easy to launch, with clear steps for downloading the pretrained weights and starting the server.

I’d also like it to work smoothly on a single GPU if possible, and if there are faster options like xformers or TensorRT, expose them as simple switches. Keep the interface friendly and minimal, and make sure the whole thing feels like a polished demo I can run locally.