AI Music Prompts: How to Write Them + 12 Prompt Ideas

A good AI music prompt tells the generator what the song is about, how it should feel and what it should sound like — in concrete words, not adjectives alone. The formula is simple: subject, genre, mood, instruments, tempo and voice. Below you'll find how each part works, more than ten example prompts you can adapt, and the mistakes that make results sound generic.

The anatomy of a prompt that works

Think of a prompt as a brief you'd give a session band. They need to know the story and the sound. In SongAI's Simple mode you have up to 3,000 characters for this; in Custom mode the style field takes up to 200 characters and your lyrics carry the story.

1. Subject — what or who the song is about. This is the part people skip most, and it matters most for lyrics. "A road trip with my best friend in a car with no air conditioning" gives the AI real material.
2. Genre — the musical family. Pick it in the genre selector (22 options) and, if you like, refine it in words: "dusty country with a little blues", "bedroom pop".
3. Mood — the emotional colour. Choose one of the 16 moods, then add nuance: "nostalgic but not sad", "confident, a bit cocky".
4. Instruments — the textures you want to hear: "warm Rhodes piano", "fuzzy bass", "string section", "808s and hi-hat rolls".
5. Tempo and energy — "slow and spacious", "mid-tempo groove", "fast, driving", "starts quiet and builds to a huge last chorus".
6. Voice — male, female or auto in the selector, plus delivery in words: "breathy and close", "raspy", "spoken-word verses, sung chorus". Prefer no vocals? Use the Instrumental switch.

You don't need all six every time. But a prompt with a clear subject and two or three sound details will almost always beat a long list of adjectives.

Example prompts you can adapt

Use these as starting points. Swap the details for your own.

A warm acoustic pop song about my grandmother's kitchen on Sunday mornings: the radio, the smell of fresh bread, her humming off-key. Nostalgic, gentle fingerpicked guitar, soft female vocal.

Upbeat indie rock about quitting a job I hated and riding my bike home in the rain, laughing. Jangly guitars, driving drums, a shout-along chorus.

Lo-fi hip-hop instrumental for late-night studying. Dusty vinyl crackle, mellow jazz piano chords, soft boom-bap drums, slow and steady.

A dark, cinematic synthwave track about driving through an empty neon city at 3 a.m. Pulsing analog bass, gated drums, distant male vocal, mysterious.

Feel-good reggae about a lazy beach day with friends, flip-flops, a cooler that's always empty. Offbeat guitar skanks, deep bass, relaxed male vocal.

An epic orchestral piece for a fantasy adventure trailer: low brass, pounding percussion, choir swells, builds from quiet tension to a triumphant climax. Instrumental.

Smooth R&B slow jam for our fifth anniversary. Mention our first apartment with the broken heater and dancing in the kitchen. Romantic, silky female vocal, soft keys.

Aggressive phonk with a cowbell melody, distorted 808s and a dark, hypnotic loop. Short chant-like male vocal hooks.

A melancholic blues song about missing the last train home and walking the city alone. Slide guitar, slow shuffle, gravelly male vocal.

Euphoric EDM anthem about the moment the festival lights come on. Big supersaw build, a drop with a chopped vocal hook, energetic female vocal.

Peaceful ambient soundscape for meditation: soft pads, slow evolving textures, distant chimes, no drums. Instrumental.

A funny country song about my dog, Biscuit, who steals socks and is terrified of the vacuum. Twangy guitar, upbeat, playful male vocal.

Notice the pattern: a real situation, a couple of concrete details, then sound. That's the whole trick.

Describe the sound, not the artist

Real artist names are blocked in SongAI, and so are well-known copyrighted lyrics. Requests like "in the style of" a famous singer won't go through (and if a request is rejected, your credit is refunded). This is deliberate: the goal is original music, not imitation.

The good news is that describing the sound usually works better anyway, because it tells the AI exactly what you liked about that artist. Translate a name into its ingredients:

• Instead of a famous pop star → "glossy 80s-inspired synth-pop, punchy drums, big layered chorus, confident female vocal".
• Instead of a famous rock band → "crunchy guitars, anthemic chorus, raw male vocal, stadium energy".
• Instead of a famous rapper → "laid-back West Coast-style groove, G-funk synth lead, smooth flow".

Era, scene, instruments, production style and vocal texture — that's the vocabulary that gets you the sound you're hearing in your head.

Common mistakes (and quick fixes)

• Only adjectives. "A beautiful, emotional, amazing song" gives the AI nothing to grab. Fix: add a subject and one physical detail.
• Conflicting instructions. "Calm and relaxing heavy metal with aggressive screaming" pulls in two directions. Fix: decide on one main feel; contrasts work better as structure ("quiet verses, explosive chorus").
• Repeating the settings. Writing "sad pop song" after selecting Pop and sad wastes space. Fix: use the words for what the selectors can't say.
• Too many details for one song. Ten memories crammed in become a list, not a song. Fix: pick the two or three strongest.
• Names that get mispronounced. Fix: spell them the way they sound — "Shivon" instead of "Siobhan".
• Changing everything between attempts. Fix: change one thing at a time so you learn what works.

When to switch to Custom mode

If you keep liking the music but not the words, move to Custom mode. Put a short sound description in the style field (up to 200 characters) and write or paste lyrics with tags like [Verse] and [Chorus]. Our song structure guide shows how to lay them out, and the lyrics generator can give you a structured draft to edit. Already have finished lyrics? Go straight to lyrics to song.

Not sure which genre or mood to pick? Browse the genres and moods pages for a feel of each.

Try it

Take one of the prompts above, swap in your own subject and details, and paste it into the AI song generator. Each generation gives you two versions, so you can hear right away which direction your prompt pulls in.