Stability AI · Sound effects

Stable Audio 3.0 SFX

Commercial use: conditions applyStability AI Community License Agreement

Soundbox's default sound-effect model; the prompting advice in the README is written for it. It runs from a quantised GGUF on audio.cpp's C++/ggml runtime instead of torch, which gives a much smaller footprint and no Python stack. The cost is that the model reloads on every generation, so a batch of short clips pays that overhead on every clip. It responds to physical description (material, action, space), not to mood words.

stabilityai/stable-audio-3-small-sfx on Hugging Face audio.cpp

The licence

Summarized from the linked source documents, last checked 25 August 2026. Every claim links to its source; anything unconfirmed is marked unconfirmed.

Stability AI Community License Agreement

Commercial use: conditions apply

Commercial use, including selling what you make, is free while you or your organisation have under USD $1,000,000 in annual revenue. Above that you need an Enterprise licence from Stability AI. The threshold applies to your total annual revenue from all sources.

Conditions

Section III: "If at any time You or Your Affiliate(s), either individually or in aggregate, generate more than USD $1,000,000 in annual revenue" you must obtain an Enterprise License.
Stability's original weights are gated on Hugging Face. Soundbox downloads audio.cpp's GGUF conversion from an ungated repo, so you need no Hugging Face account or token. The licence still applies to the GGUF weights.
The models embed a pre-trained T5Gemma text encoder redistributed under the Gemma Terms of Use, which also apply however you obtain the weights. The Gemma terms were not reviewed for this page, so what they add is not documented here.

What about the audio you generate?

Output is covered by a separate clause, in your favour. Section IV(c)(iii): "As between You and Stability AI, You own any outputs generated from the Models or Derivative Works to the extent permitted by applicable law." Stability's FAQ says you may therefore use those outputs at your discretion. The ownership clause does not mention the $1M threshold; the threshold governs your right to run the model. That reading combines the two clauses; the licence does not state it in those words.

Based on the licence text as of the date below. Licences change; the links carry the authoritative version.

Specs

Transcribed from the shipped catalog, audiocpp_models_catalog.json — that file is authoritative if a number here disagrees with it.

Vendor: Stability AI
Catalog id: audiocpp-stable-audio-3-sfx
audio.cpp loader family: stable_audio
Can do: Generation
Weights repo: audio-cpp/audio.cpp-gguf
Weights file: Stable-Audio-3-Small-SFX-GGUF/stable-audio-3-small-sfx-q8_0.gguf
Account needed to download: No
Max clip length: 120s
Sample rate: 44.1kHz stereo
Approx. VRAM: 2.5 GB
Download: ~2 GB
In the picker by default: Yes
Run end-to-end here: Yes

Getting it

There is no per-model install. One native binary runs all of these, so getting a model is a download of its weights and nothing else — no virtualenv, no torch pin, no CUDA wheel to match against your driver.

From the CLI

soundbox model install audiocpp
soundbox pull audiocpp-stable-audio-3-sfx

The binary is fetched from audio.cpp's GitHub releases the first time you ask for it and cached under your own data directory. Soundbox ships no native binaries and vendors none.
Weights land in <models dir>/audiocpp/audiocpp-stable-audio-3-sfx/ rather than in the shared Hugging Face cache, so a download stays scoped to the model you asked for.

Worth knowing

Soundbox downloads audio.cpp's GGUF conversion from an ungated repo, so no Hugging Face account or token is needed. Stability's original weights are gated, and the Community License still applies to the GGUF.
Defaults are 8 steps at CFG 1.0, and the catalog caps CFG at 3.0. This is a few-step model; high guidance makes the output worse without making it follow the prompt more closely.
No negative prompt. No family on the audio.cpp path exposes one, so the composer does not show that field for this model.

Sources

Primary sources: the model's own repository, the project it came from, and the conversion audio.cpp actually downloads. Never invented — an entry with nothing findable says so instead.

Licence details on these pages are summaries of the linked source documents, last checked 25 August 2026. They are not legal advice, and Soundbox is not affiliated with any model's authors — weights download from Hugging Face under the user's own account and terms. Company and model names identify whose work it is, and belong to their owners. The linked licence text is authoritative; commercial use should be checked against it, per model, before anything ships.