Stability AI released Stable Audio 3.0, a family of four music and sound generation models trained entirely on licensed data from partnerships with Universal Music Group and Warner Music Group. Three are open-weight on Hugging Face: Small SFX and Small (both 0.6B parameters) for sound effects and full music composition on consumer hardware, and Medium (2B parameters) for longer, more complex tracks up to 6 minutes 20 seconds. The architecture runs on a novel semantic-acoustic autoencoder (SAME) that compresses audio into a compact latent space, cutting generation to under two seconds on an H200 GPU and a few seconds on a MacBook Pro M4. Small generates up to two minutes of music, a 10x jump from 11 seconds in the previous version. LoRA fine-tuning (for customizing the model on your own audio) and audio inpainting (editing specific sections mid-track) are both supported.
Under the Stability AI Community License, users own their outputs and can commercialize them freely (organizations above $1M revenue need an Enterprise license). Most open music models either restrict commercial use or train on unlicensed data, which creates legal risk. A podcaster or indie game developer can download the 0.6B model, fine-tune it on their own audio library, and generate royalty-free music offline on a laptop. No API costs, no copyright uncertainty.
Music AI is splitting into two camps: proprietary platforms like ElevenMusic and Google’s Lyria 3 that control the pipeline end-to-end, and open-weight models you can run anywhere. Stable Audio 3.0 is the strongest entry in the open camp so far, and it comes with clear legal footing.
Sources:
- Meet Stable Audio 3.0 (Stability AI)
- Stable Audio 3 Collection (Hugging Face)
- Stable Audio 3 Research Paper
- SAME Autoencoder Paper
- Stable Audio 3.0 Day-0 Support (ComfyUI Blog)
Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service
Reuse
Citation
@misc{kabui2026,
author = {{Kabui, Charles}},
title = {Stable {Audio} 3.0: {Open-Weight} {Music} {Generation}
{Trained} on {Fully} {Licensed} {Data}},
date = {2026-05-23},
url = {https://toknow.ai/posts/stability-ai-stable-audio-3-open-weight-music-generation/},
langid = {en-GB}
}