Edge Dance

EDGE, short for Editable Dance Generation from Music, is the Stanford research project presented at CVPR 2023. Its official project page links an accessible PyTorch implementation. The model pairs music features from Jukebox with a transformer-based diffusion model to create dance motion, with joint constraints, in-betweening and continuation. It suits researchers and technical animation teams evaluating music-conditioned motion rather than users expecting a maintained, hosted choreography service.

The project describes stitching five-second motion clips with temporal consistency constraints. The repository documents custom audio input, reusable music features and motion export through FBX for Blender rendering and retargeting. These are published workflows, not results from a generation or compatibility test. The separately linked web demo currently resolves to unrelated game marketing, so that link should not be treated as an available EDGE application. The official code repository remains the verified public entry.

The documented environment recommends Linux, Python 3.7 or later, PyTorch 1.12.1 and CUDA 11.6. It calls for at least 16GB of system memory per GPU and high-end NVIDIA hardware with at least 16GB of graphics memory. Windows is not officially supported. Those historical requirements do not establish compatibility with every current library or operating-system release. The authors explicitly describe a research implementation that generally will not receive long-term updates after release.

The repository's MIT licence permits software use and redistribution subject to retaining its copyright and permission notice, and provides no warranty. No paid hosted plan or fixed service fee was established. Compute, storage and any external services remain separate costs. Check licences for checkpoints, dependencies, datasets, music and character assets before commercial work; the repository licence alone does not settle all those rights. No checkpoint was downloaded, software installed, music uploaded or model run during this review.