Harmonai
Stability AI's open-source music AI lab behind the Dance Diffusion audio models.
About
Harmonai is the open-source generative audio research lab backed by Stability AI. Its flagship release, Dance Diffusion, was among the first diffusion models for raw audio, generating music clips by learning to destroy and recover audio samples, with public checkpoints trained on piano, vocals and glitch music datasets.
Unlike closed music generators, Harmonai publishes training notebooks, pretrained checkpoints and code that run locally or in Colab, including the sample-generator repo with models at 16 kHz and 48 kHz and the oobleck neural audio codecs. It targets producers, researchers and hobbyists who want to experiment with and fine-tune diffusion audio rather than rent a black-box generator.
What it can do
- Open-source Dance Diffusion audio generation models
- Unconditional generation, style transfer and clip interpolation
- Pretrained checkpoints at 16 kHz and 48 kHz
- Oobleck neural audio codecs
- Fine-tuning notebooks for custom audio libraries
Setup
Clone the Harmonai sample-generator repo or open the Colab notebooks; download a pretrained Dance Diffusion checkpoint to generate locally.
Prerequisites
Python environment with PyTorch; GPU recommended for local generation.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 126; Trending score 72/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.