Dance Diffusion
Open-source diffusion for raw-audio music generation, variation and interpolation.
About
Dance Diffusion was Harmonai's 2022 answer to the question of what Stable Diffusion would look like for sound. The Stability AI-backed lab, founded by Zach Evans, released it as an open diffusion model operating directly on raw audio rather than MIDI, so it can synthesize nearly any sound. Users generate clips from scratch, upload their own audio and regenerate variations of it, or interpolate between two tracks to morph one into the other.
The first trained model, J Mann, learned from thousands of songs by Jonathan Mann, and the design invites anyone to train their own model on music or samples they own. Quality was openly rough, grainy and warbly by the team's own admission, and the Stable Audio line has since taken the spotlight, but Dance Diffusion's checkpoints and Colab notebooks remain available and it stays historically important as the first major open music diffusion model.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No formal benchmarks; qualitative demos only.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: public beta Sep 2022 from Stability AI-backed Harmonai, open checkpoints and notebooks, documented train-your-own workflow. Strengths: fully open, raw-audio diffusion, variation and interpolation features. Docked for: (1) early-generation audio quality; (2) superseded in practice by Stable Audio; (3) no formal evaluations. License: open-source (Harmonai).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Harmonai (Stability AI) beta, Sep 2022
- · Raw-audio diffusion; generate, variate, interpolate
- · J Mann model trained on Jonathan Mann's song catalog
- · Train-your-own on owned samples; Colab notebooks
- · Superseded in practice by Stable Audio line; Harmonai logo staged