Stable Audio 3

Last updated: 16 October 2026

Dots can run Stable Audio 3 by Stability AI on your computer. These are the key points of its terms. You accept them before your first run, and they add to our Terms. The model's own documents always win: Stability AI Community License, Acceptable Use Policy and model card.

Capabilities and intended use

Stable Audio 3 makes music and sound effects from text, and can edit or continue a piece of audio. Its open models (Small SFX, Small and Medium) run on your own computer; Medium makes the most musical and longest tracks, up to about six minutes.

Out-of-scope uses

Anything Stability AI's Acceptable Use Policy forbids, including breaking the law or other people's rights (such as copyright and privacy), any sexual content involving minors, non-consensual intimate content, content showing sexual acts, hate, harassment, encouraging self-harm, impersonation, misinformation, and getting around safety measures. You must be 18 or older, or the legal age where you live if that is higher. Outputs may not be used to train other foundation AI models. Stability AI's policy and license win if anything here differs.

Training data

Trained on about 1.3 million recordings: audio licensed from AudioSparx, and Creative Commons recordings from Freesound (CC0, CC BY and CC Sampling+). Before training, Stability AI checked the music in the data and removed copyrighted content it found. The Freesound recordings are credited on Stability AI's attributions page, linked from the model card.

Safety

Stability AI says the models are trained responsibly on licensed data; the model card lists no further safety tests. Its Acceptable Use Policy also applies when you run the model yourself, and Stability AI may end access for people who break it. Because the weights are open, what you make with them is your responsibility.

Risks and limitations

It is new technology. Results can't be predicted and may not match your prompt or sound the way you expect. Prompts are in English. The largest model (Large) is not open and doesn't run in DotStories.

License and outputs

You own the audio you make, as far as the law allows, and may use it commercially. The license is free for people and organizations under US$1M yearly revenue; above that, you need a license from Stability AI, and commercial users must register with Stability AI. The model's text encoder (T5Gemma) is also under Google's Gemma Terms of Use. License: Stability AI Community License.

Reporting and contact

Report harmful, illegal or policy-violating results to us (Terms, section 9). We pass them on to Stability AI when needed. More about the model: model card.

Contact

Nestor.gg
Mercatorplein 36D, 1056 CL Amsterdam, the Netherlands
KvK (Dutch Chamber of Commerce): 94770662 · VAT: NL005107992B34
Email: info@dotstories.com

See also: Terms · Privacy · Cookies · Code of Conduct