Short answer
Write one or two sentences with four parts: a genre or palette, two or three instruments, a tempo or feel, and the job in the video ("under a voiceover", "lifts at the product reveal", "ends on a hit"). Generate, compare the two versions against the edit, and change one thing at a time.
01
Write the brief before the prompt
Before you type a genre, answer four questions about the video. They take a minute and they decide most of what the prompt should say.
- What is the video? A travel reel, a product ad, a tutorial. The kind of video sets the expectations of the audience.
- What does the music do? Carry momentum through fast cuts, sit under a voice, build to a reveal, or close on an end card.
- Where is the voice? If someone talks for most of the video, the music needs space: sparse, no busy lead melody, steady level.
- How long, and does it change? A 15 second ad needs a hook and an ending. A three minute vlog needs something that stays even.
Then turn the answers into a palette: a genre, two or three instruments, a tempo or feel, and the job. That is the whole prompt.
"Lo-fi, chill, jazz, hip-hop, relaxing, study" names six moods and asks for none of them. "Warm lo-fi hip-hop at 85 BPM with soft Rhodes and upright bass, mixed low under a voice" asks for one track.
02
Example one: a talking vlog
The brief: a three minute day in the life vlog, someone talking to camera most of the time, light and friendly.
Warm lo-fi hip-hop at 85 BPM with dusty vinyl crackle, soft Rhodes chords and a lazy upright bass, mixed low so a voice sits on top.
Why it works: one genre, three instruments, a tempo, and a job. "Mixed low so a voice sits on top" tells the model the music is a bed, not the lead. Set the length to match the vlog, or leave Duration on Auto.
03
Example two: a 15 second product ad
The brief: a short ad for social, product reveal in the middle, logo on the last two seconds, no voiceover.
Sleek minimal electronic at 120 BPM with a deep pulse, crisp clicks and a bright synth stab on the product reveal, ending on a clean hit.
Why it works: an ad lives on structure. The prompt names the moment that matters (the reveal) and how the track ends (a clean hit), which is what lets the music land on the end card instead of fading out under it. Set Duration to the length of the cut.
04
Example three: a narrated explainer
The brief: a two minute explainer with narration from start to finish over motion graphics.
Soft minimal electronic with plucked synths and a gentle kick at 100 BPM, low in the mix under a voiceover explaining software steps.
Why it works: the prompt asks for less. Under constant narration, the right music is the one nobody notices: steady, sparse, no peaks. If you have the video, this is also where keeping music under a voiceover starts.
05
Text to Music or Video to Music
The same prompt works in both, but it does a different job.
- Text to Music makes music from the prompt alone. The prompt sets everything, so it carries the structure: "ends on a hit", "builds after the intro". Choose Auto, a length from 30 seconds to 3 minutes, or a custom length from 5 seconds to 8 minutes. Each generation gives two versions, and the music is always instrumental.
- Video to Music takes the length, pacing and scene changes from the video itself. There the prompt is the direction: style, instruments and mood. Leave it blank and the music is matched to the video on its own.
For a track with distinct parts, turn on sections in Text to Music and give each section (intro, verse, chorus, outro and so on) its own prompt and length; the track is as long as the sections added together.
06
Revise one variable at a time
When a result is close but wrong, change one thing and generate again. Changing the genre, the tempo and the instruments together makes it impossible to tell which change helped.
| What you hear | Change this one thing |
|---|---|
| Too busy under the voice | Add "sparse, no lead melody" or remove an instrument |
| Too slow for the cut | Raise the tempo (for example 90 to 110 BPM) |
| Right mood, wrong sound | Swap one instrument: piano for guitar, strings for synth pads |
| No moment for the reveal | Name it: "a bright stab on the reveal", "a drum fill into the chorus" |
| Fades out instead of ending | Say how it ends: "ends on a clean hit", "cut to silence" |
The Enhance button rewrites a short prompt into a fuller one without spending credits, and you can undo the rewrite. It is a good way to see which details a prompt is missing.
07
Save what works
When a prompt lands, keep it with the project: the prompt, the length and which of the two versions you used. A prompt that worked for one vlog is the best starting point for the next one. The prompt library is 300 of those starting points, grouped by the kind of video.
FAQ
Frequently asked questions
Related
Keep reading
Start from a prompt