Inference uses your device's WebGPU hardware. No prompt or audio is uploaded.
STABLE AUDIO OPEN SMALL · LOCAL WEBGPU
Describe it.
Hear it.
Create original foley, ambience, impacts and production sounds. Your prompt and generated audio stay on this device.
PrivateUp to 11 seconds44.1 kHz stereo
Try:
Checking WebGPU compatibility…
Ready to generate
The model downloads once on first use (about 2 GB), then remains in your browser cache.
Best for ambience, foley, percussion and production elements—not realistic speech.
Stable Audio Open Small is used under the Stability AI Community License.