Live-Portrait.org

Live-Portrait.org is an animate face tool that turns one still portrait into a moving, expressive video. Live-Portrait.org animates a single still photo using motion from a driving video, with sliders for eye and lip control. It is the web playground for the open-source Live Portrait framework, supporting cross-reenactment, stylized portraits and video-to-video face editing.

Live-Portrait.org is the browser front end for Live Portrait, a video-driven portrait animation framework built by researchers from Kuaishou Technology, the University of Science and Technology of China, and Fudan University. The premise is simple: give it a single still photo of a face and a short driving video, and it maps the motion from the video onto the photo. The output keeps the identity of the source image while borrowing the expressions, head pose and lip movement of the reference clip. The system rests on an implicit keypoint framework trained on roughly 69 million high-quality frames. Features are extracted from the source image and the driving video, the motion is transferred through keypoints, and an optimized decoder synthesizes each frame quickly (the project reports 12.8ms per frame on an RTX 4090 under PyTorch). Two modules sit on top of that core: stitching, which keeps the animated region aligned with the rest of the frame, and retargeting, which lets you nudge eye openness and lip position with simple scalar inputs instead of re-shooting anything. In practice the workflow is three steps. You upload a portrait with a clear, forward-facing face, choose or upload a driving video, then press animate and wait. Most users can leave the configuration defaults alone, though the control sliders are there for anyone who wants to push a smile wider or close the eyes slightly. Beyond simple reenactment from a driving clip, the same pipeline supports cross-reenactment (moving motion from one person onto a different portrait), video-to-video face editing, and stylized inputs such as oil paintings, sculptures and 3D renders. The underlying models can also be fine-tuned for animal faces, cats and dogs among them. It sits at the research-grade and open-source end of the portrait animation market. The inference code and weights are public on GitHub, so teams that want to self-host or fine-tune have a route, while the free playground covers quick one-off tests. Compared with polished commercial avatar studios, Live-Portrait.org is leaner: it does not bundle voice cloning, script writing or template libraries, and it expects you to supply your own driving footage.