MusePose

Why we included this project

MusePose turns a single reference photo of a person into a video where that same character follows a given pose sequence, so a static portrait can be made to dance or move in sync with a driving skeleton. It builds on the AnimateAnyone line of work but adds a pose-alignment step that fits arbitrary dance footage to arbitrary reference images, which removes a lot of the fiddly setup that similar models demand. Both inference and training code are released, and the project ships as a self-contained Python pipeline using diffusers, mmpose, and DWPose, so expect to download weights and run it on a GPU rather than call an API. Teams making virtual hosts, character animation, or any content that needs a still figure to move on cue will find this one of the more practical open implementations for pose-controlled human video around.