Stability AI releases a new audio model that can create six-minute songs

Stability AI, the institution down Stable Diffusion, is releasing a caller household of audio models, called Stability Audio 3.0. The apical exemplary tin make professional-grade euphony of much than six minutes long, the institution claimed.

The institution is releasing 4 caller models nether the Stable Audio 3.0 name: tiny SFX (459M parameters), tiny (459M parameters), mean (1.4B parameters), and ample (2.7B parameters). The duo of tiny models is suitable for on-device dependable and euphony procreation of up to 2 minutes.

Both mean and ample models tin make afloat compositions of 6 minutes 20 seconds agelong that tin support philharmonic operation and melodic tone. This is much than treble the magnitude of what Stable Audio 2.0, released successful 2024, was susceptible of generating.

Stability AI is making tiny SFX, small, and mean models disposable with unfastened weights for anyone to usage and modify. In 2024, the institution released Stable Audio Open, which allowed for euphony procreation of up to 47 seconds. The caller household of models is simply a large measurement up from the erstwhile unfastened versions.

Image Credits: Stability AIImage Credits:Stability AI

The ample exemplary is disposable lone done the API and self-hosting paid services. Plus, companies with much than $1 cardinal successful gross would request to get an endeavor license.

Many companies, including Google and ElevenLabs, are releasing models and tooling astir euphony generation. However, arsenic Suno and Udio’s ongoing tribunal battles person proved, licensing of information and partnerships with euphony labels could go a cardinal portion of the semipermanent endurance of these services.

Last year, Stability AI inked deals with Warner Music Group and Universal Music Group to make models and euphony instauration tools. The institution said that its latest acceptable of audio models is built connected afloat licensed data.

The AI startup is processing a caller suite of products for nonrecreational musicians, but didn’t springiness much details connected its features. Ethan Kaplan, erstwhile main integer serviceman astatine Universal Audio and Fender, is joining the institution to pb Stability’s nonrecreational euphony offering.

A fig of AI companies are trying to bolster their credentials by hiring euphony execs. Earlier this year, Suno hired erstwhile Merlin CEO Jeremy Sirota arsenic main commercialized officer. ElevenLabs has besides hired Derek Cournoyer from indie euphony steadfast Kobalt arsenic a strategy pb for its euphony business.