EDGE

About EDGE

EDGE, developed by Jonathan Tseng, Rodrigo Castellon, and C. Karen Liu from Stanford University, is a revolutionary tool that generates editable dances from any music input. Introduced at CVPR 2023, EDGE has set a new standard in the field of dance generation.

Background and Development

EDGE was born out of a desire to create a tool that could generate realistic, physically-plausible dances from any music input. The team behind EDGE paired a transformer-based diffusion model with Jukebox, a strong music feature extractor, to achieve this goal. The result is a tool that not only generates high-quality dances but also offers powerful editing capabilities.

Core Features and Capabilities

EDGE's unique features include joint-wise conditioning, motion in-betweening, and dance continuation. These features allow users to generate dances that are not only realistic but also customizable to their specific needs. EDGE also uses music embeddings from Jukebox to understand and create dances for in-the-wild music samples.

User Experience

Users have found EDGE to be a powerful and user-friendly tool. Its ability to generate dances of any length by imposing temporal constraints on batches of sequences has been particularly appreciated. Additionally, EDGE supports arbitrary spatial and temporal constraints, making it a versatile tool for various applications.

Applications and Use Cases

EDGE can be used in a variety of scenarios, from generating dances for music videos to creating choreographies for dance performances. Its ability to generate dances subject to joint-wise constraints also makes it a useful tool for dance training.

Impact and Future Outlook

EDGE has made a significant impact in the field of dance generation, with human raters strongly preferring dances generated by EDGE over those of previous work. As the team continues to refine and expand EDGE's capabilities, its impact is only expected to grow.