AI Music Generator
Describe the feeling and get a track back at the length you need. Made for video: beds that sit under narration rather than songs that compete with it.
- Mood, genre, and tempo from a prompt
- Generated to the length of your cut
- Instrumental beds that leave room for a read
- Same library, same credits as your video
How to score a video with generated music
Scoring is a shape problem before it is a genre problem. Decide what the music has to do under the picture, then describe that.
- 01
Describe the function, not just the genre
'Warm, sparse, slow build, nothing above a whisper under the voice' is a brief. 'Lo-fi hip hop' is a shelf. Say what the track is for and the result is far more likely to fit the cut.
- 02
Set the length to the edit
Generate to the duration you actually need. A bed cut down from a longer piece usually ends mid-phrase, and fading it out is the sound of a track that was not written for this video.
- 03
Listen against the picture
Play it under the cut before judging it. A track that is interesting on its own is often the wrong one — under narration, the good bed is usually the one you stop noticing.
What generated music solves
No licence to keep track of
Music you generate under your plan comes with the same commercial terms as everything else you make here — one thing to check rather than a per-track licence to file.
Written to length
Ask for the duration your cut needs and the arrangement is built to it, so it ends rather than fades.
Beds that leave room
Instrumental, mid-range-light beds under a voice over are a different request from a song, and describing it that way gets you a track a narrator can survive.
Variants for testing
The same brief three ways, so an ad can test tone the way it tests copy.
Alongside the footage
Generate picture, narration, and score from one balance and find them in one library, ready to cut.
Nothing anyone else is using
A generated bed is not the stock track currently under ten thousand other videos on the same platform.
What people score
Short-form social
Fifteen to sixty seconds with a lift where the payoff lands, generated to that exact shape.
Product videos
A restrained bed that supports a voice over without pulling attention off the screen.
Ads
Several tonal directions for the same script, so the choice is made on evidence rather than taste alone.
Titles and stings
Short pieces for an opener, an outro, or a transition, at the length the edit leaves for them.
Ambient loops
Background texture for a long-running scene, stream, or installation.
Podcast and video chapters
Consistent chapter beds so a series sounds like a series.
Briefing music the way an editor would
Most disappointing generated music comes from a prompt that names a genre and stops. Genre tells the model what instruments to reach for and nothing about what the track has to do. An editor briefing a composer would say where the piece starts, where it lifts, what has to be audible over it, and what should be happening in the last two seconds — and those four facts are exactly what improves a generated result.
Density is the setting most worth stating explicitly. Music under narration needs to be sparse in the range a human voice occupies, and busy percussion or a strong melodic line in that range will fight every word. Asking for 'sparse, low and high, nothing in the middle' sounds technical, but it is the difference between a bed and a distraction.
Length deserves the same care as it does in video. Generating to the duration of the cut means the arrangement resolves where the picture resolves. Generating something longer and fading it under the last shot is the most audible tell that music was chosen rather than written, and it is avoidable now that generating to length costs nothing extra.
Finally, judge in context and be willing to run again. A bed that seems bland alone is frequently the right one once there is a voice and a picture on top; a track that grabs you in isolation often turns out to be the reason the audience stopped listening to the script. Two or three generations against the actual cut settle it faster than any amount of listening on its own.
AI music FAQ
- Can I use generated music commercially?
- Yes, under the terms of service, on a paid plan — the same terms as generated video and images. There is no separate per-track licence to manage.
- How long can a track be?
- Long enough for typical short-form and mid-length video; the available durations are shown in the form. For longer material, generate sections and cut them together.
- Can I generate vocals?
- The studio is aimed at instrumental beds for video. For anything that has to speak, use the voice over tool, which is designed for intelligibility rather than musical phrasing.
- Can I match an existing track?
- Describe what you like about it — tempo, instrumentation, density, mood — rather than naming the song. Naming a specific commercial track is neither reliable nor something to build a campaign on.
- Is it free?
- Yes, within your free credits, from the same balance as image and video generation.
Where to go next
Describe a mood and hear it at your length
Free credits on sign-up, tracks generated to the duration of your cut, and one place for picture and sound.