How to write AI music prompts that sound like what you imagined

The five things every song prompt needs

Most weak prompts describe a feeling ("a sad song") and nothing else. The model then guesses everything else, and the guess is usually the most average version of that feeling. Give it anchors instead:

• Genre, ideally with one qualifier: "90s alt rock", "bedroom pop", "cinematic orchestral".
• Mood and energy: "bittersweet but hopeful", "tense and driving", "lazy Sunday".
• Two to four instruments: "fingerpicked acoustic guitar, upright bass, brushed drums".
• Voice: gender, texture, and delivery, or "instrumental": "breathy female vocal, close and intimate".
• Tempo or feel: a BPM like "92 BPM", or "half-time", "swung", "four-on-the-floor".

A template you can reuse

Put the five anchors in one sentence, most important first. Anything that is not an anchor, such as your life story or a list of ten genres, tends to blur the result.

[genre] song about [topic], [mood], [instruments], [voice], [tempo]

Indie folk song about moving to a new city, bittersweet but hopeful, fingerpicked guitar and soft piano, warm male vocal, 96 BPM

Examples by genre

Copy one of these into the Grok Music composer in Simple mode, then change one detail at a time to learn what each word does.

• Pop: "Upbeat synth-pop about a summer crush, bright plucks and punchy drums, confident female vocal, big singalong chorus, 118 BPM"
• Hip hop: "Boom bap hip hop, dusty piano loop, heavy kick and snare, laid-back male rap verses, sung hook, 88 BPM"
• Rock: "Driving alt rock, crunchy guitars, live drums, raspy male vocal, explosive final chorus, 140 BPM"
• Lo-fi: "Instrumental lo-fi beat for studying, warm Rhodes chords, vinyl crackle, soft swung drums, 75 BPM"
• Cinematic: "Instrumental epic trailer music, low strings, taiko drums, brass stabs, slow build to a huge climax"
• R&B: "Smooth modern R&B, 808 bass, airy synth pads, silky falsetto male vocal, late-night mood, 90 BPM"

Simple mode or Advanced mode?

Simple mode takes one description and writes everything, including lyrics. It is the fastest way to explore ideas. Keep it under 500 characters.

Advanced mode splits the prompt into three fields: style and arrangement, title, and lyrics. Use it when the words matter. The style field should describe only the sound; put anything that should be sung in the lyrics field.

Fixing common problems

When a result is close but not right, change one thing and generate again. These fixes cover most cases:

• Too generic: add a specific instrument and a tempo.
• Wrong energy: say where the energy goes, e.g. "quiet verses, explosive chorus".
• Unwanted sounds: in Advanced mode, list them under Exclude styles, e.g. "autotune, EDM drop".
• Wrong voice: describe texture and range ("low, husky alto") rather than naming a real artist, which is not supported.
• Messy mix of genres: keep one main genre and at most one influence.

Compare takes before rewriting

Each generation returns separate takes, and they can differ a lot. Before you rewrite a prompt, listen to all of them. If one is close, it is often faster to extend it or make a cover of it than to start again.