AI Music Prompt Guide
How to write text-to-music prompts that actually get the sound you want — the ingredients that matter, vague-versus-specific examples, and tips for Suno, Udio, MusicGen and Stable Audio.
The anatomy of a good prompt
Vague prompts get generic music. The fix is to specify the things a session musician would ask about — genre, mood, instrumentation, tempo and production — in plain language.
| Ingredient | Examples |
|---|---|
| Genre / style | synthwave, drill, bossa nova, ambient techno, gospel |
| Mood / energy | melancholic, euphoric, tense, laid-back, cinematic |
| Instrumentation | analog pads, fingerpicked guitar, 808s, string section, gospel choir |
| Tempo | 120 BPM, slow and spacious, driving, half-time |
| Key / tonality | minor key, dark, bright major, modal |
| Era / production | 70s tape warmth, lo-fi, modern and polished, vinyl crackle |
| Structure | intro–build–drop, loopable 8-bar phrase, sparse verse into a big chorus |
Vague vs specific
- "a happy song"
- "electronic music"
- "euphoric melodic house, 124 BPM, warm analog pads, plucky lead, side-chained bass, festival energy"
- "lo-fi hip hop, 78 BPM, dusty Rhodes chords, soft boom-bap drums, vinyl crackle, rainy-night mood"
Tips by tool
- Suno / Udio — split style and lyrics. Use section tags in the lyrics like [Verse], [Chorus], [Bridge], and keep the style box to genre + mood + key instruments.
- MusicGen / Stable Audio — purely descriptive, no lyrics. Lead with genre and instrumentation; Stable Audio in particular responds well to an explicit BPM.
- Everything — generate several takes per prompt and iterate one variable at a time.
Common mistakes
- Piling in contradictory genres — pick a lane.
- Forgetting tempo and mood, the two biggest levers on feel.
- Over-long prompts; a focused sentence or two beats a paragraph.
Put these into practice with MusicGen, or match a tempo with the tap-tempo tool.