This platform does not export a bare melody line. It generates complete tracks, and the tune inside that track is something you steer rather than extract. This page explains how to get the melody you want without expecting note-level or MIDI control.
- By description. Register, contour, mood and era all move the tune: "rising synth lead in a high register", "descending minor piano motif", "call-and-response between guitar and synth".
- By tempo and genre. A bpm figure plus a genre tag fixes the rhythmic frame the melody sits in.
- By reference audio. The music workbench accepts an uploaded audio file (up to 20MB) alongside images and video, runs media analysis on it, and uses the result to steer generation. That is the closest thing here to "make something that moves like this".
- By iteration. Regenerate with one changed variable, or extend the track and describe the next section's melodic direction.
- By vocal direction. If the melody is sung, the vocal style field shapes who is singing and how.
- Open the AI Music Generator and sign in — generated tasks belong to your account.
- Choose the mode that matches the material. Text to Music for an idea in your head, media upload when you have a reference clip, Lyrics to Music when the words already fix the rhythm of the tune.
- Describe the melody, not just the genre. Add register, shape and feel to the style string.
- Upload a reference if you have one. Run the media analysis step, then check that the analysis captured what matters before you generate.
- Pick the model and read the credit cost shown next to it.
- Generate and listen to the hook. If the tune is right but the arrangement is wrong, change only the arrangement words.
- Extend or regenerate for a second section. Use Extend when you want a variation that grows out of the first.
- Export the finished track from the result panel.
- No MIDI or sheet export. The output is audio, so you cannot open the melody in a notation app or drag it into a piano roll.
- No solo or mute per instrument while it plays. There is no mixer in the workbench; audio separation (two or multiple stems) gives you parts after the fact, not during.
- No note-by-note correction. Fix a melody by re-prompting, not by editing pitches.
- No guaranteed commercial rights. Rights are not automatically granted and depend on the selected model provider, your plan, your inputs, applicable law and third-party rights — see the Terms of Service.
- No fixed budget. Each attempt consumes credits at the cost shown for the selected model.
- Prompting the word "melody" and nothing else. A melody needs a genre, a tempo and a register to be interesting.
- Changing five things between tries. You cannot attribute the improvement, so keep the change list to one item.
- Uploading a long reference clip. A short, clear phrase steers better than a full song, and the form enforces its own size limit.
- Expecting a DAW-ready part. Plan to either use the audio as it is or re-perform the tune yourself.
No. The task returns a mixed track. If you need the melodic content isolated, run vocal separation to remove the vocal, or use the multiple-stem option and see what the separation gives you.
Not as a transcription. You can upload an audio reference for the model to analyse, and it will steer the generation, but the platform does not convert a hummed line into notes you can edit.
Generate the section, then use Extend with the source track and describe the repeat — for example "restate the opening motif on strings". The result is a new section derived from what you already approved.
It is the same generator, so the cost is the per-task credit cost of the model you select, displayed before you submit. Current packs and plans are on Pricing.
Open the AI Music Generator →
