Brief

Researchers have developed MotionDreamer, a diffusion-based framework capable of generating skeletal animations for 3D rigged shapes using 2D video as guidance. This method is designed to be category-agnostic, meaning it can animate a wide variety of 3D models regardless of their topology or morphology. To address the lack of diverse training data, a large dataset of approximately 20,000 3D models with animations was curated. MotionDreamer integrates texture and semantic attributes into its skeletal joint representations to accurately map visual motion cues to different 3D structures, achieving state-of-the-art results in 4D asset generation. AI