menyifang/MIMO

Decomposes video into spatial components (character, motion, scene) to enable independent control over each attribute during synthesis. Built on Stable Diffusion with motion modules and pose guidance, it supports animating static character images with 3D motion data or driving videos while generalizing to novel characters and scenes. Provides both local PyTorch inference and a Gradio web interface, with pre-trained weights available via HuggingFace and ModelScope.

1,575 stars. No commits in the last 6 months.