Audio Gen · Model Guide · 4 min read
Stable Audio Open
Open-weights sound effects and audio stingers — drum hits, foley, textures, and ambient beds from a text prompt.
| Developer | Stability AI |
|---|---|
| Availability | Open weights |
| Commercial license | Stability AI Community License: free for individuals and organizations under $1M annual revenue, including commercial projects at that scale; larger organizations need a paid license. Generated outputs are usable under the license — read the terms for the revenue threshold that applies to you. |
Basic Information
Stable Audio Open is Stability AI's compact open-weights model for generating short audio: sound effects, foley, percussion loops, synthesizer textures, and ambient beds up to tens of seconds. It is trained on Free Music Archive and other openly licensed audio, which is what makes its licensing posture defensible, and it integrates with the standard local diffusion tooling.
It is not a music-composition model in the Suno sense — no full songs with vocals — but for generating usable, license-clean sound design assets on your own machine it is the reference open option.
Best use cases
- Game and app sound effects generated to spec (whooshes, impacts, UI clicks)
- Ambient loops and texture beds for video and installations
- Foley-style experimentation without sample-library licensing costs
- Fine-tuning on your own sound library for a custom palette
Computational Requirements
Small: roughly 1–2 billion parameters, comfortably running on 4–8 GB GPUs and workable on CPU for short clips.
| Configuration | Memory |
|---|---|
| Model size | ~1.2B parameters |
| Comfortable VRAM | ~4–8 GB (less with quantized builds) |
Commercial license
Stability AI Community License: free for individuals and organizations under $1M annual revenue, including commercial projects at that scale; larger organizations need a paid license. Generated outputs are usable under the license — read the terms for the revenue threshold that applies to you.
Compare the cost of using it
Curious how a hosted model’s bills add up against local hardware? The Local vs Cloud panel does the math for your workload.
More guides: back to Audio Gen · visit the blog