Stable Audio 3.0
#15 en Génération musicale IAStability Ai · v3.0 · 5× · vu le 30 juin 2026
Stable Audio 3.0 is a family of generative audio models (Small SFX, Small, Medium, Large) released by Stability AI in May 2026, based on latent diffusion with a novel semantic-acoustic autoencoder. The models generate music and sound effects from text prompts of variable length up to approximately 6 minutes 20 seconds and support audio inpainting as well as continuation of existing clips. Three of the four models (Small SFX, Small, Medium) are available as Open Source weights on Hugging Face, while the flagship Large model is only accessible via the Stability AI API or Enterprise Self-Hosting. All models were trained exclusively on licensed or Creative Commons data (including AudioSparx and Freesound); voice cloning and vocal generation are not supported.