How to Write a Better AI Music Prompt

A useful music prompt is less like a shopping list and more like a concise creative brief. It gives the model enough constraints to make coherent decisions while leaving room for a musical performance.

The six-part prompt

Build your prompt from the elements that matter most:

1. Mood and setting — What should the listener feel, and where does the music seem to live?
2. Genre and era — Name a musical vocabulary, not a specific living artist to imitate.
3. Tempo and energy — Slow, mid-tempo, driving, restrained, explosive, or a BPM range.
4. Instrumentation — Choose a few defining sounds and say which one leads.
5. Structure — Mention the arc: sparse intro, rising verse, wide chorus, short bridge, quiet ending.
6. Voice and production — Vocal character, language, recording texture, space, and mix priorities.

A complete example

Intimate late-night alternative R&B at a slow tempo. Warm electric piano leads, supported by soft sub bass, brushed percussion, and distant guitar harmonics. Start minimal, build tension through the verse, then open into a wide but restrained chorus. Breathable low-register vocal, close-mic production, subtle tape texture, no aggressive drums.

This works because every detail supports the same emotional world. “Distant guitar harmonics” and “close-mic vocal” say more than a list of ten unrelated instruments.

What to avoid

• Contradictory directions such as “minimal” and “maximal orchestral wall” without explaining a transition
• Long lists of genres that do not share a common rhythm or sound palette
• Instructions about every second of the track
• Using a celebrity's name as a shortcut for musical qualities
• Vague praise words such as “amazing” or “viral” without audible characteristics

Revise one dimension at a time

After listening, decide whether the main issue is composition, sound, energy, voice, or structure. Change that dimension first. If you rewrite everything at once, you lose the ability to learn which instruction changed the result.

The Agent is useful here: tell it what you heard and what felt wrong, then ask for a tighter second prompt. Keep the strongest phrases, remove instructions the music already understands, and make each iteration more specific rather than simply longer.