stabilityai / stable-audio-3-small-music

Overview

Stable Audio 3 Small Music is a text-to-music model from Stability AI. It uses latent diffusion to denoise a compact music representation, then decodes it to audio. You get a ready-to-use stereo track at 44.1 kHz, which helps you draft cues without recording.

What you can build

• Short instrumental cues for video edits, ads, and social posts
• Loopable background music for games, apps, and menus
• Genre and mood explorations for writing sessions or storyboards
• Draft music beds for podcasts and trailers (instrumental-style prompts work best)
• Fast A/B variations of the same idea by changing tempo, instruments, or mood

Inputs

• A text description of the music you want, written in English. Include genre, instruments, mood, and tempo (BPM).
• A target length in seconds for the generated track. The Small Music checkpoint tops out at 120 seconds.
• An option to expand a short prompt into a longer, more detailed one before generation.
• A setting that controls how many denoising iterations the model runs. Fewer iterations produce faster drafts.
• A setting that controls how strongly the result follows your text description.

Outputs

• One downloadable audio file of the generated music, typically delivered as an MP3.
• Metadata that echoes the final prompt used for generation.
• A numeric seed value you can reuse to reproduce the same result later.

Recommended settings

• Start with 8 denoising iterations for this distilled Small Music checkpoint.
• Keep prompt-following strength around 1.0. Higher values usually don’t help on distilled checkpoints.
• Turn on prompt expansion when your prompt is a single sentence or shorter.
• Use 20 to 60 seconds for quick iteration. Use up to 120 seconds for a full cue.

Limitations

• Output length is capped at 120 seconds for Stable Audio 3 Small Music.
• It’s not designed for speech or voice generation.
• Prompts written in English work best. Other languages often degrade quality.
• Very short or vague prompts can produce generic results. Add BPM, instruments, and mix notes.
• If your request implies a specific copyrighted song or a living artist’s exact style, results may be inconsistent or blocked.

Safety & compliance

• Use is governed by the Stability AI Community License. Organizations over $1M annual revenue may need an Enterprise license.
• Don’t generate content that violates rights, privacy, or platform content rules.
• Some deployments apply an automated safety checker and may refuse disallowed prompts.
• You are responsible for how you use and distribute generated audio in your project.