2026-09-27

AI music prompt examples: a template plus 6 briefs

A good AI music prompt reads like a brief: mood, genre, BPM, key, instruments, one moment, and era. Here is a template and six examples to adapt.

5 min readSume

All posts

A good AI music prompt is a short brief, not a genre tag. Name the emotion, the genre, the tempo in BPM, the key, two to four instruments with their texture, one moment in the arc, and the era or production style, and end background music with “Instrumental, no vocals.” Sume's docs use this one: “Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes chords, brushed boom-bap drums, a muted trumpet answer at 0:10. A 30-second track. Instrumental, no vocals.”

The template and the rules come from Sume's Music 1.0 and Music Router docs, read on 2026-09-27. The docs call the brief's details creative directions, not guaranteed output settings, so listen to every result. How to call the API is in Music generation API.

What should an AI music prompt include?

Fill one slot per axis, in this order, and end with the vocal clause. The docs recommend this scene-specific brief as the way to avoid generic musical choices.

Close with “Instrumental, no vocals.” Add “no spoken word” only when the music will sit under narration.

[Emotion], [genre or lineage], [tempo] BPM, [key and mode].
[2–4 instruments, each with a texture].
[Arc: how it starts, one named moment at a timestamp, how it ends].
[Era or production]. [Length: "A 30-second track."]
Instrumental, no vocals.

What are some AI music prompt examples?

Six briefs written to that template. Swap the details, keep the shape.

Podcast narration bed: “Calm, curious documentary bed, 90 BPM, F major. Felt piano, soft upright bass, light shaker. Sparse and steady; a warm string pad enters at 0:30 and the piano thins out for the last ten seconds. 2020s, clean and close. A 1-minute track. Instrumental, no vocals, no spoken word.”
Intro sting: “Bright, confident indie-pop sting, 120 BPM, A major. Palm-muted electric guitar, handclaps, a glockenspiel hook. A four-bar build with a snare fill at 0:06, ending on one ringing chord at 0:08. Modern, tight and punchy. An 8-second track. Instrumental, no vocals.”
Product ad: “Playful, upbeat funk-pop, 112 BPM, E minor. Slap bass, muted guitar chops, tight live drums, bright horn stabs. The drums drop out at 0:12 and the full band returns at 0:14 for the reveal. 2010s, polished and loud. A 30-second track. Instrumental, no vocals.”
Trailer cue: “Tense, brooding cinematic cue, 142 BPM half-time, D minor. Low string ostinato, taiko hits, a distorted synth drone. Builds for 40 seconds, stops dead at 0:40, then one final hit at 0:42. 2024 hyper-clean. A 45-second track. Instrumental, no vocals.”
Study ambient: “Hushed, weightless ambient, 60 BPM, E phrygian. Granular pads, soft felt piano, distant tape hiss. No drums; a low cello note enters at 1:00 and holds to the end. 1990s cassette warmth. A 2-minute track. Instrumental, no vocals.”
Fusion theme: “Proud and nostalgic gugak fusion, 96 BPM, G major with a lydian lift. Gayageum plucks, warm strings, soft hand drums. A gentle opening; the strings swell at 0:20 and the gayageum takes the melody to the end. 2024 hyper-clean. A 1-minute track. Instrumental, no vocals.”

How do I prompt contrasting scenes in one project?

Change three things between scenes: the broad genre family, the tempo (at least 12 BPM apart), and the lead instrument. A hushed chamber-folk scene at 76 BPM led by nylon guitar and a driving synthwave scene at 118 BPM led by an analog synth differ on all three. For one consistent score, keep them the same.

How do I set the length or leave something out?

Both go in the prompt text. The request has no length, seed, temperature, or guidance setting, and it refuses a non-empty negative prompt.

From Music 1.0 and Music Router, read 2026-09-27.
You want	What the docs say
A set length	Say it in the prompt (“a 2-minute track”) or use section markers such as `[0:00-0:30] Intro: …`. `duration` and `duration_seconds` are rejected.
To exclude something	Write it into the prompt, e.g. “no vocals, no spoken word”. A non-empty `negative_prompt` returns HTTP 400 with `negative_prompt_unsupported`.
Repeatable output	There is no seed, temperature, or guidance parameter.
A long, detailed brief	`prompt` takes 1–5,000 characters.
Visual steering	An optional public HTTPS `image_url` as visual conditioning; see Image to music AI.

What if the result misses the brief or the prompt is rejected?

Listen before you use it: the docs call the axes creative directions and say to verify the generated audio. result.lyrics, when present, carries the model's lyrics or section map. It may describe tempo and structure, but the docs call it model-reported metadata, not an audio measurement.

On a policy rejection, the docs say to revise the flagged content while keeping the musical brief, and not to strip the request to a generic bed.

How do I send a music prompt to an API?

Put the brief in prompt on POST /v1/music-router/generate. API pricing lists music generation at $0.125 per audio generation, plus a 5.5% agent fee by default, and the docs say a longer prompt doesn't change the price. If you pin a model id, GET /v1/music-router/models lists its pricing. To cut a short sting to an exact length, see AI jingle generator from text.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: narration-bed-001" \
  -d '{
    "prompt": "Calm, curious documentary bed, 90 BPM, F major. Felt piano, soft upright bass, light shaker. Sparse and steady; a warm string pad enters at 0:30 and the piano thins out for the last ten seconds. 2020s, clean and close. A 1-minute track. Instrumental, no vocals, no spoken word."
  }'