Moore-AnimateAnyone

Why we included this project

This is one of the more complete community implementations of the AnimateAnyone approach. Feed it a single character reference image and a driving pose sequence, and it generates a video of that character following the poses, so a drawn or photographed person can be animated without re-drawing every frame. The maintainers also added a face reenactment mode that uses facial landmarks from a driving video to steer head pose, eye blink, and mouth movement while keeping the source identity, which makes it handy for talking-head or avatar work. The repository ships inference code and pretrained weights along with full training scripts, so you can run it as is or fine-tune on your own footage. Just keep expectations in check: the authors call it a preliminary build that gets roughly 80% of the way to the original paper's results, and known rough spots include background artifacts on clean-reference images and jitter on static scenes.