Stable Audio Open

Stable Audio Open is Stability AI's open-weight text-to-audio model, generating variable-length stereo audio of up to 47 seconds at 44.1 kHz from a text prompt. Architecturally it pairs an autoencoder that compresses waveforms into a latent space, a T5-based text encoder for conditioning, and a diffusion transformer that generates in that latent space, totaling around one billion parameters. Unusually for the field, training data provenance is documented: roughly 486,000 recordings drawn from Freesound and the Free Music Archive, all under CC0, CC BY, or CC Sampling+ licenses, with content detection used to filter suspected copyrighted material before training. The model is aimed at sound effects, field recordings, foley, and production elements rather than full songs; it cannot generate realistic vocals, was trained only on English descriptions, and is uneven across music genres. Weights are hosted on Hugging Face under the Stability AI Community License, with larger commercial use requiring a separate license, and the model attracts sound designers and researchers who want fine-tunable open audio generation.