Stable Audio 3

Stable Audio 3 is a family of text-to-audio models for generating music and sound effects. The useful part is the range of workflows around the model: you can generate a new clip from a prompt, restyle an existing recording, regenerate a section, or continue audio beyond its original ending.

That makes it more interesting than a one-shot prompt-to-sample tool. For example, you can start with a short synth idea, ask for a longer dream-pop instrumental, then use inpainting or continuation when only part of the result needs another pass.

The reader-facing workflow is the public Stable Audio 3 Space. It exposes Medium, Small Music, and Small SFX in one Gradio interface. Medium is the tracked general model here, while the smaller variants are useful when you want a lighter music-only or sound-effects path.

Start with the Hugging Face Space and choose the variant that matches the job. Use Small SFX for a short effect, Small Music for a lightweight music test, or Medium when you want the broader long-form audio path.

For the first prompt, avoid a generic genre label. Write one concrete production brief: style, instruments or sound source, mood, tempo, and duration. For example, test a 30-second cinematic neo-soul groove, a station train arrival with horn, or a synth-pop instrumental with a clear BPM. Listen for whether the timing, texture, and structure match the prompt, not only whether the clip sounds polished.

If the browser test is promising, move to the model page and the Stable Audio 3 GitHub repo. The local workflow includes Python, CLI, and Gradio paths, plus audio-to-audio editing and inpainting examples. Expect Medium to be a real GPU setup; the model gate, CUDA, and Flash Attention requirements are part of the practical cost.

The public Space is the fastest test path, but local use is not frictionless. You need to accept the Hugging Face model gate, and Medium is a GPU workflow with CUDA and Flash Attention setup details. For lighter experiments, start with the small variants before committing to the full local path.

• Draft background music for short videos, demos, and moodboards.
• Generate sound effects for games, UI prototypes, or video edits.
• Extend a short audio idea into a longer rough track.
• Regenerate one section of a clip instead of starting over.
• Compare Small Music, Small SFX, and Medium on the same audio brief.

Try the demo

View model page