How to Generate Music with AI

Write an AI music prompt using genre, mood, instruments, vocals, and length. Learn how to create instrumentals, songs, and video soundtracks.

A track from a sentence

Gab AI can compose original music from a written description. Same composer as chat and images, same credit balance. You are not looking for a sample pack — you are describing a finished cue: what genre it is, how it should feel, what is playing, and whether anyone is singing.

Describe a record, not a vibe word

“Make a sad song” is a coin flip. “A slow 70s soul ballad, upright piano and brushed drums, female vocal, about leaving a small town, 2 minutes” is a brief a musician could take.

The five parts of a music prompt

You do not need all five every time. Genre and mood get you most of the way. Instruments, vocals, and length are how you stop getting a generic trailer cue.

See the difference

Same request. One is a shrug. The other is a brief.

A prompt you can paste

This one uses all five parts. Change the genre and it still holds together.

Pair it with video

Generate the picture or clip first, then ask for a track that matches the pacing. Or reverse it: write the music, then generate a video that can cut on the downbeats. The music lesson after video exists for this reason — they are one workflow, not two hobbies.

Royalty and taste

These tracks are original generations, not ripped hits. Do not prompt “in the exact style of [living artist]” if you plan to publish. Name the era, the instruments, and the feeling instead.

Describe a record, not a vibe word

“Make a sad song” is a coin flip. “A slow 70s soul ballad, upright piano and brushed drums, female vocal, about leaving a small town, 2 minutes” is a brief a musician could take.

  1. Genre — Name the tradition: lo-fi hip hop, chamber folk, 80s synthwave, gospel, film score. “Cinematic” alone is not a genre.
  2. Mood and tempo — Hopeful, tense, tender, triumphant — plus a pace. “Mid-tempo, 95 BPM” or “dirge-slow” is enough.
  3. Instruments — List what should be audible. A Rhodes, a nylon guitar, a drum machine, a string quartet. If you do not name them, the model will invent a safe default.
  4. Vocals — or none — Instrumental, hummed, male lead, choir, rap verse. If you want words, say the subject. If you want a bed for a video, say instrumental.
  5. Length — Thirty seconds for a bumper. Two to three minutes for a full cue. Say it, or you will get whatever the model’s default is.

Royalty and taste

These tracks are original generations, not ripped hits. Do not prompt “in the exact style of [living artist]” if you plan to publish. Name the era, the instruments, and the feeling instead.