Stanford-TML/EDGE

Leverages a transformer-based diffusion model with Jukebox music feature extraction to generate physically-plausible dance sequences from audio input. Provides fine-grained editing capabilities including joint-wise conditioning and in-betweening, with a novel Physical Foot Contact (PFC) metric for evaluating motion quality. Integrates with PyTorch3D and Hugging Face Accelerate for training, and supports FBX export for 3D animation workflows in Blender or Mixamo.

552 stars. No commits in the last 6 months.