Back to Gap Map Model components / Image, video, 3D & music generation

stable-audio-tools

Stability AI
open source / Overall score: 1.6

stable-audio-tools is Stability AI's open training and inference library for audio generation models. It trains and runs latent diffusion models such as Stable Audio Open, the Dance Diffusion waveform models and the audio autoencoders they decode with, from JSON model and dataset configurations, with LoRA fine-tuning and a Gradio interface for trying a trained model.

Scored as the library. The Stable Audio weights it loads are their own entry, stable-audio, under Stability's community license; the newer Stable Audio 3 models ship their own pipeline in a separate repository.

Openness

5 high confidence
5.0
license
MIT (OSI; the stable-audio-tools code. The third-party code it vendors, under LICENSES/, is MIT as well where read)
source
public (Stability-AI/stable-audio-tools)
core features withheld
no — no ee, enterprise, commercial or proprietary path; Stability sells its largest audio model through a separate API and enterprise offering, not as an edition of this library

The library is MIT-licensed and fully public, training code included. Stability AI's paid audio offering is a hosted model, not a closed edition of this code.

Adoption

1 high confidence
1.0

Adoption is measured as PyPI downloads of the stable-audio-tools package, published by Stability AI. The README's main install path clones the repository, which is not counted.

Capability

2 high confidence
2.0

stable-audio-tools trains and runs more than one family of audio generation models, as AudioCraft does, and like it covers audio alone.

Verified 2026-11-12, except some scores, still awaiting re-confirmation