Stability AI released Stable Audio 3.0 this week, a set of four music generation models that create audio from plain text descriptions. You type something like "upbeat jazz with piano, 90 seconds, for a product video" and the model produces the track. Three of the four models are free to download and run on your own hardware, including on a phone.
The family covers four different use cases. The two smaller models handle sound effects and short music on any consumer device. The medium model produces tracks up to six minutes and twenty seconds. The large model, reserved for enterprise API access, targets high-volume use cases like music platforms and content services. The jump in length is significant: the previous open version topped out at 47 seconds. Now you can generate a complete background track in one go.
The technical upgrade is real, but the bigger story is the legal one. Stability AI trained these models on a licensed audio library from production company AudioSparx and open-license recordings from Freesound. That is a deliberate contrast to competitors Suno and Udio, which built popular tools by training on commercial recordings without permission. Both faced major label lawsuits. Warner Music settled with Suno in November 2025, and Universal Music settled with Udio in October 2025. Sony is still litigating, and a court ruling expected this summer could set a precedent affecting every AI music company.
For businesses that want to use AI audio commercially, training data provenance now carries legal weight. Courts have consistently ruled against AI defendants in cases where training data was pirated, and have been more willing to consider fair use arguments for legitimately acquired data. Stability AI also offers legal indemnification under its enterprise license, meaning the company would defend you if a copyright claim arises from using the tool. That is an unusual and commercially significant guarantee.
The commercial terms are straightforward. Anyone can download the small and medium models and use them freely. If your organization earns more than $1 million a year, you need an enterprise license to commercialize the output. Users below that threshold own what they generate under a community license.
The company also has formal partnerships with Universal Music Group and Warner Music Group, though those label catalogs were not used to train this specific release. The music industry veteran Ethan Kaplan, former chief digital officer at Universal Audio and Fender, is joining Stability AI to lead a professional musician product line that has not yet launched.
Stability AI's own history adds context. The company went through severe financial difficulties in 2024 after founder Emad Mostaque departed under pressure. A new CEO, a restructured balance sheet, and roughly $80 million in fresh investment stabilized operations. It is a leaner company now, with about 190 employees, rebuilding around audio and enterprise contracts rather than trying to compete on every AI front.
For non-technical operators who produce any kind of content, the practical consideration is this: AI-generated background music and sound effects are now a real cost-saving option, and Stable Audio 3.0 is one of the cleanest legal paths to get there. The tool will not write a hit song, but it will produce functional, original audio for internal videos, event presentations, apps, or client-facing content without a licensing fee or a session musician.