StabilityAI announced AI Music Generator Harmonai based on Dance Diffusion Model

In Brief

Harmonai brings open-source audio models to help shape the future of music

Dance Diffusion aims to make music production more approachable and enjoyable for everyone

The Trust Project is a worldwide group of news organizations working to establish transparency standards.

With Harmonai, a cleverly called new method for, you guessed it, creating an audio out of nothing, StabilityAI’s researchers have made a big advancement in the field of AI audio-generation. The outcomes are astounding and diverse, and they are all just a little bit unsettling.

StabilityAI announced AI Music Generator Harmonai based on Dance Diffusion Model
Image generated by Stable Diffusion

What is Harmonai?

Harmonai is a community-driven organization that publishes open-source generative audio tools to increase everyone’s access to and enjoyment from music composition. To put it simply, it generates never-before-heard sounds in a process called diffusion, in which random white noise is refined into sounds based on a pre-trained model that contains millions of parameters that were calculated during its training process to encapsulate the characteristics of the source material it was trained on.

The goal of the deep learning research lab Harmonai is to produce open-source generative audio models, support researchers and developers through compute funding and a community and offer new creative tools to artists and audio professionals. The goal of Harmonai is to make music production more approachable and enjoyable for everyone by developing open-source audio synthesis and manipulation tools.

What’s Dance Diffusion?

Imagine Stable Diffusion only producing sound instead of images. Dance Diffusion is a collection of publicly available generative music diffusion models. The initial models generate 1-3 seconds of audio and can also be used to interpolate and style transfer lengthier audio recordings.

Is it therefore similar to text-to-image AI for audio?

In spirit, yet with some significant variances. There isn’t a text entry form where you can type “snare drum” and press GO, thus this isn’t a text-to-audio or text-to-speech generator. Never say never, of course—this project is still in its early stages—but for now, it’s up to you, the user, to supply a pool of audio files (be they snare drums or whatever it is you wish to generate variations of) to train the model on. Alternatively, you can use one of the 6 pre-trained models already present in the Dance Diffusion notebook to get you started—and there’s plenty of fun to be had playing with them alone!

Yep. Harmonai is Dadabots + more amazing people + 4000 GPUs. In their YouTube channel, Dadabots’ artists already use AI generated music:

Freshly generated AI audio samples by Harmonai

Follow the community discord channel to listen to more audio files.

Interested in reading more? Here is some additional news to check out:


Any data, text, or other content on this page is provided as general market information and not as investment advice. Past performance is not necessarily an indicator of future results.

Damir Yalalov

Damir is the Editor/SEO/Product Lead at He is most interested in SecureTech, Blockchain, and FinTech startups. Damir earned a bachelor's degree in physics.

Follow Author

More Articles