Stable Audio 3 Medium

Stable Audio 3 Medium is an open audio model from Stability AI with 1.4 billion parameters that creates music and sound effects from a text description. Per the README, it also edits existing audio by prompt (audio-to-audio), continues and fills in tracks, and can be fine-tuned with LoRA. Clips run up to 380 seconds, stereo, 44.1 kHz. The README does not mention vocals or singing, so we do not claim them. The same family has the smaller Small-Music and Small-SFX models (433M parameters, up to 120 seconds, can run on CPU); the largest Large version is available only through an API and is not an open model. Released on 2026-05-17. The weights are on Hugging Face, and you download them after accepting the terms (gated access, which we count as open). The license is the Stability AI Community License. Per the official license page, it covers researchers, developers, small businesses and creators with less than 1 million US dollars in annual revenue; organizations above that need a paid Enterprise license. The FAQ there says that, as between you and Stability, you own the outputs, and that the models and their outputs cannot be used to create new foundation models. We took these terms from the official page and did not read the license file itself, which is behind the acceptance step. The authors offer an official demo Space on Hugging Face; code is on GitHub and the paper is on arXiv. The same developer's image models are in our catalog: Stable Diffusion. Date of check: 2026-10-18.