Audio PromptMonster

How to write prompts for Stable Audio 3

Last checked 4 October 2026Not affiliated with Stability AI

Stable Audio 3 makes instrumentals: it has no lyrics input and no intelligible vocals. Begin the prompt with TrackType: Music, VocalType: Instrumental,, then the genre, the main instruments and rhythm, the mood, the arrangement and the production. Set the length as a setting, up to 380 seconds through the API.

What does Stable Audio 3 read?

Stable Audio 3 reads one text prompt and a length. It makes music, single instruments and sound effects, but no words: Stability says it can make vocal-like textures, never intelligible singing, lyrics or speech.

Stability's technical report says the hosted models were trained with a track-type prefix, and recommends adding it to every music prompt because it noticeably improves quality.

WhatValueSource
Vocals and lyricsInstrumental only: no lyrics input, no intelligible singing or speechStability AI, prompt guide, 4 October 2026
Recommended prefixTrackType: Music, VocalType: Instrumental, for music (TrackType: SFX, for sound effects)Stable Audio 3 technical report, 4 October 2026
Prompt10,000 characters on the API; the model reads about the first 256 tokensStability AI API reference, 4 October 2026
Length1 to 380 seconds on the API (190 if not set); the web app offers fixed lengths and ExtendStability AI API reference, 4 October 2026
TempoWritten as "N BPM" in Stability's own examplesStability AI, GitHub prompting guide, 4 October 2026
Things to avoidNo negative prompt on the hosted APIStability AI API reference, 4 October 2026
Modelstable-audio-3 on the API; open-weight small and medium models tooStability AI, model overview, 4 October 2026
OwnershipYou own what you make and may use it commercially (Community License)Stability AI, licensing, 4 October 2026
APMonster's promptThe prefix, then comma-separated fields, under 1,500 charactersAPMonster's own practice

What does Stability AI say a prompt should contain?

Stability's prompt guide says to name the genre first, then the main instruments and rhythmic elements, then the mood, the arrangement and the production. Tags such as TrackType: Music or Genre: Funk are optional labels, and its training prompts join fields with commas, with the tempo written as "N BPM".

The API takes up to 10,000 characters, but the model reads only about the first 256 tokens, so the words that matter go first.

How does APMonster write a Stable Audio prompt?

APMonster starts every Stable Audio prompt with TrackType: Music, VocalType: Instrumental,, then writes comma-separated fields in the order Stability's guide gives: the genre, the main instruments and rhythm, the moods, the arrangement and the production, with the tempo as "N BPM", as in Stability's own examples. What matters most comes first, because the model reads only about the first 256 tokens, and the prompt stays under 1,500 characters.

Because the model sings nothing, APMonster's lyric writer, LyrAssist, writes a vocal guide for Stable Audio instead of lyrics: melody shape, phrasing and delivery for a singer to perform over the track.

How do you tell Stable Audio what to avoid?

Stability's hosted API takes no negative prompt, and its open-weight package applies one only to base models. The documented way is to describe what you want: "a clean, warm mix" rather than naming distortion.

APMonster does the same, inside the prompt: a brief that avoids heavy distortion asks for clean, warm production instead.

How does Stable Audio take song structure?

Stable Audio takes no section tags. APMonster describes the song's arc in a sentence in the prompt and gives the length to set: up to 380 seconds through the API. Stability's report notes that quality and prompt adherence drop at the longest durations.

The example song's structure, as APMonster writes it for Stable Audio 3 
Opens with a soft instrumental intro, builds into a steady verse, surges into a strong chorus, drops to a steady verse, surges into a strong chorus, drops to a soft bridge, surges into a full-force chorus, and ends on a very soft instrumental outro.

This sentence goes into the prompt, and the duration setting is the song's length, 2:47.

One brief, written for Stable Audio 3

Every guide on this site writes the same song for its platform. This is the shape APMonster asks for on Stable Audio; the wording of a generated prompt varies from one generation to the next.

Prompt
TrackType: Music, VocalType: Instrumental, Genre: dream pop, Instruments: chorus-soaked electric guitar, warm analog synth pads, soft drum machine, round bass, Moods: wistful, hopeful, slow build into each chorus, clean, warm late-1980s production, 92 BPM
The arc (added to the prompt)
Opens with a soft instrumental intro, builds into a steady verse, surges into a strong chorus, drops to a steady verse, surges into a strong chorus, drops to a soft bridge, surges into a full-force chorus, and ends on a very soft instrumental outro.
Duration setting
2:47

What changed for Stable Audio 3, and why

  • The prompt opens with TrackType: Music, VocalType: Instrumental,, the prefix Stability recommends.
  • The fields follow in Stability's order, separated by commas, with the tempo written "92 BPM".
  • The breathy female vocal was dropped: Stable Audio 3 makes instrumentals. LyrAssist writes a vocal guide for a singer instead of lyrics.
  • The key was left out, and the structure became a sentence, with the length as a setting.
  • Things to avoid were said positively: clean, warm production in place of heavy distortion, and a soft drum machine in place of trap hi-hats.

Frequently asked questions

Can Stable Audio 3 make songs with vocals?

No. Stable Audio 3 makes instrumentals, single instruments and sound effects. It can make vocal-like textures, but no intelligible singing, lyrics or speech.

How long can a Stable Audio 3 track be?

Up to 380 seconds through the API (190 seconds if no length is set). Stability notes that quality and prompt adherence drop at the longest durations.

Does Stable Audio 3 have a negative prompt?

Not on the hosted API. Describe what you want instead, such as a clean, warm mix; APMonster writes things to avoid that way.

What is the track-type prefix?

A tag the hosted models were trained with: TrackType: Music, VocalType: Instrumental, for music and TrackType: SFX, for sound effects. Stability recommends adding it to every prompt.

Can I use Stable Audio 3 tracks commercially?

Stability says you own what you make and may use it commercially under its Community License; an Enterprise licence applies above a revenue threshold.

Sources