Source: singularity gallery - https://gall.dcinside.com/mgallery/board/view/?id=thesingularity&no=913108

https://huggingface.co/huaichang/PersonaLive

Hugging Face

*Intro written in 3-line Boomer/Zoomer summary style

[3-Line Summary] * Real-time Generation: Create a video that instantly mimics the user's expressions and movements using just one photo. * Streaming Optimized: Reduces latency drastically using 'Micro-Chunk Streaming' technology, ready for immediate deployment in live broadcasts. * Low Barrier to Entry: Works on consumer GPUs (12GB VRAM) without expensive gear, and supports popular tools like ComfyUI.

What if you could create a virtual avatar that copies your expressions and actions in real-time, all from just one picture? The new AI model 'PersonaLive,' released on Hugging Face, is here to make that happen. This model takes existing 'Talking Head' technology a step further, moving beyond simple lip-syncing to deliver high-quality human character animation optimized for live streaming. ◇ "One Photo and I'm a Vtuber" PersonaLive's core strength is its 'Single-shot' processing capability. Instead of complex 3D modeling or lengthy training, you just input one photo of the person you want, and that person comes to life, matching your movements via your webcam. This makes natural communication possible in fields like virtual YouTubing, online classes, or video conferencing, without needing to reveal your identity. ◇ Technical Innovation for Live Broadcasts Existing video generation AIs were too slow for live broadcasting. But PersonaLive introduced a unique technique called 'Autoregressive micro-chunk streaming.' By chopping the video into tiny units and processing them in real-time, it enables seamless, 'infinite-length' video generation. This minimizes common issues like character distortion or freezing, even during long broadcasts. ◇ "My PC is Enough"... Excellent Accessibility The most welcome news is the running environment. Unlike massive models from Tencent or Google that require servers costing tens of thousands of dollars, PersonaLive runs smoothly on a single consumer-grade graphics card (GPU) with 12GB VRAM. It also officially supports 'ComfyUI,' widely used in the AI image generation community, allowing average users to easily install and try it out right away. 'PersonaLive' is currently free on Hugging Face and GitHub, allowing anyone to create their own 'living persona' using their photo.