Back to blog

How to Write a Better AI Music Prompt

Updated AI Musician TeamAI Musician Team

Build stronger music briefs with three practical prompts, a weak-to-strong example, lyric preparation, and a focused listening-and-revision process.

Overhead view of a music notebook, pencil, sticky notes, headphones, and part of a keyboard on a desk

A better music prompt starts with a decision: what job should this piece do? Background music beneath speech, a pop chorus, and an exploratory game theme need different arrangements. Start there before collecting genre labels.

Adapt the three briefs below, then listen. Describing a tempo or length does not guarantee an exact BPM, duration, or edit point.

AI Musician Music Generator interface with the prompt and generation controls
The Music Generator interface is where you review the musical brief and available settings before submitting a task. This screenshot shows the interface, not a newly generated result.

Build a brief around audible choices

Choose compatible decisions from these six dimensions rather than filling every category with adjectives.

DimensionA useful instructionWhy it helps
PurposeBackground for a spoken travel diaryEstablishes how much space to leave
MoodReflective, with quiet optimismGives the piece an emotional center
Musical vocabularyAcoustic folk with a relaxed pulseConnects rhythm and sound palette
Lead and supportFingerpicked guitar leads; soft bass supportsMakes the arrangement hierarchical
DevelopmentSparse opening, warmer middle, gentle endingDescribes an arc without a second-by-second script
Voice and textureInstrumental; close, natural recordingClarifies the foreground and production feel

Put the strongest constraint first: room for narration, for example, or the melodic and dynamic contrast of a chorus.

Change vague praise into a musical brief

Weak: “Amazing cinematic music for my video. Emotional, epic, relaxing, with lots of instruments.”

The model must reconcile competing energy levels and guess what “amazing” means. Replace those judgments with decisions you can hear:

Stronger: “Reflective acoustic instrumental for a travel diary with spoken narration. Fingerpicked guitar leads over a restrained bass line and light percussion. Keep the opening spacious, introduce a warm pad in the middle, and finish gently. Avoid big drum hits and a busy lead melody beneath the voice.”

Now you can judge whether the guitar leads, the middle develops, and the narration has room.

Three prompts to adapt

1. A podcast or travel-film bed

Warm acoustic instrumental for a thoughtful spoken story. Fingerpicked guitar, rounded bass, and light brushed percussion. Unhurried pulse, steady energy, and plenty of space in the midrange. Begin simply, add a soft sustained layer as the story develops, and end with a gentle release. Keep melodic phrases short and avoid dramatic drum fills.

Enable instrumental music in Music Generator. Test the output beneath actual speech: a beautiful solo listen can still compete with narration.

2. A hopeful pop song

Hopeful indie-pop song about moving to a new city while missing home. Warm lead vocal, clean rhythmic guitar, supportive bass, and a restrained live-drum feel. Make the verse intimate and let the chorus become wider and more singable without turning into a stadium anthem. Conversational words, concrete everyday images, and a brief reflective bridge.

If particular words matter, review them separately and put your chosen lyrics into the music tool's custom lyrics field. A style description does not supply the text to sing.

3. A gentle exploration-game theme

Instrumental theme for exploring a peaceful coastal village. Soft electric piano carries a simple motif, with muted plucked strings, a rounded low end, and small hand-percussion accents. A lightly moving pulse rather than a driving beat. Let the texture gradually open, then return to a quieter arrangement. No lead vocal and no sudden orchestral climax.

Inspect transitions and endings. For a seamless game loop, edit the downloaded audio in your production workflow; a prompt does not establish loop boundaries.

Separate lyrics from musical direction

Use Lyrics Generator to explore perspective, imagery, and verse–chorus structure. “A narrator who notices small signs of home in an unfamiliar city” is more actionable than “beautiful emotional lyrics.” Read the draft aloud and replace awkward lines before generating a song.

AI Musician Lyrics Generator interface for entering a songwriting idea
The Lyrics Generator provides a separate place to prepare a writing brief. The screenshot illustrates the form; it does not show proof of a completed generation.

Keep arrangement instructions in the musical brief. In custom mode, review the style and lyric fields separately so production notes are not accidentally treated as sung words.

Revise one dimension at a time

Save the previous wording and a short listening note. Then change one row, rather than rewriting the entire concept.

What you heardChange nextKeep stable
Narration is maskedAsk for fewer foreground phrasesGenre, lead instrument, pulse
Chorus feels flatAsk for broader backing layers at the chorusStory, vocal character, verse
Drums dominateAsk for lighter percussion and fewer fillsHarmony and principal melody direction
Atmosphere is too grandReduce orchestral language and describe a smaller spacePurpose and emotional tone

A new generation may vary beyond the edited instruction. Compare the overall trade-off rather than assuming the system changed only one audio stem. Agent can help rewrite your observation into a cleaner brief; your listening remains the evidence.

Questions before your next attempt

Should I specify a precise BPM or duration?

You can describe the intended pace, but verify the completed audio. Do not plan a fixed video cut around an unmeasured generated duration. Use MIDI Studio when you want to work directly with editable notes and tempo; it is an editor, not a way to recover MIDI from every generated song.

Why did the result ignore an instrument?

First remove competing lead instruments and unrelated genres. Give one sound a clear role, generate again, and compare. More text does not necessarily create stronger control.

Does public mean featured in Showcase?

No. Free-plan generated results are public; subscribers can choose private results. Showcase features supported public tasks only after administrator approval. Review the visibility notice before submitting.

Next session, adapt one example and note the first audible change to test. Continue with the music guide or Agent workflow.

Three practical guides for your next creative step.