Stable Audio

What it is

Stable Audio is a product from Stability AI for generating music and sound from text. Stability launched it on 13 September 2023. It was described as the first tool to generate 44.1 kHz music for commercial use with latent diffusion. ^[1][2]

Origins

The first version was trained on licensed content through a partnership with the AudioSparx music library. ^[2] Latent diffusion is the technique of running the diffusion process in a compressed representation rather than on raw audio. ^[1]

The free tier allowed users to make 20-second tracks, and the Pro tier allowed tracks of up to 90 seconds. ^[2]

A sampling rate is the number of audio samples taken each second. 44.1 kHz is the rate used for CD audio, and it is enough to capture the main range of frequencies that people can hear. A stereo track has two channels, so sounds can be placed to the left or right.

Main features

Stable Audio 2.0, released on 3 April 2024, generates tracks of up to three minutes at 44.1 kHz stereo. It adds audio-to-audio transformation, which lets users transform uploaded samples. ^[3]

Stable Audio 2.0 is a latent diffusion model with a highly compressed autoencoder and a diffusion transformer (DiT) in place of the earlier U-Net. It was trained on over 800,000 licensed AudioSparx files, and artists could opt out of the training data. ^[3]

Stable Audio Open, an open-weight text-to-audio model for sound design, debuted on 5 June 2024. ^[4]

Its place in the development

In March 2025 Stability optimised its audio generation model to run on Arm chips. ^[6]

Stable Audio 2.5, released on 10 September 2025, was pitched as the first audio model built for enterprise sound production. It was trained on fully licensed data. ^[5]

Stable Audio 2.5 is available on StableAudio.com, through the Stability API, through partner platforms, and on-premises under an enterprise license. Stability worked with WPP's sound branding agency Amp on brand sound identities. ^[5][7]