Extending a song continues a generated track you already have

Already have a track that's too short? Extend it. That is the job. Extend a song is one named agent capability. extend_music continues a Suno-generated track from, or near, a point you choose. It requires that track's audioId (and a model). It does not take an arbitrary MP3 URL. Asking for "another minute of something like this" through generate a song from a prompt is a new song. That generate wastes credits on a relative of the hook you already liked.

The extend is the same song, longer. A new prompt is a new song that happens to share adjectives.

What extend_music will actually continue

Reference a track you already generated with generate_music. The agent can look up the audioId from generation history. You can point at a continueAt moment — the second chorus, the bridge — and the tool writes more music from there. You get back a longer version of that track.

Required fields are audioId and model. Optional prompt, style, and title steer the continuation; they do not replace the source. If there is no audioId, there is no extend. A file you dragged in from a phone is not a Suno id.

Cost is priced per generation and shown before you confirm. You are paying for more of a track you already bought, not for a second lottery ticket.

The AI music generator is the family door. This capability is only the continue row. The editor-shaped cousin is extend a music track.

What this job is not

A from-scratch song. generate_music takes a prompt. Use it when you do not yet have a keeper. Extending a miss makes a longer miss. Generate until you have a hook worth keeping, then extend.

A cover. Create a cover of a song is cover_music. It takes an audioUrl and reinterprets style or arrangement over the same song. Cover wants a URL. Extend wants a Suno audioId. Feeding a Spotify link to extend will not work. Feeding a Suno id to cover is the wrong contract.

Stem split. Separate vocals from a song splits a generated track into vocal and instrumental using taskId and audioId. Isolate vocals from an audio clip pulls a voice out of any mix. Neither one makes the song longer. Extending because you wanted a dry vocal is a new section you did not need.

Lyrics-only. Write song lyrics is text. Extend is audio. Do not extend in order to "fix the words." Fix the words, then generate or cover.

Continue the keeper. Do not roll a new identity.

The expensive loop is: generate a 30-second hook you like, then prompt a "full version" as a new generate_music call. You now have two songs. The second one usually drops the melody you were extending. extend_music exists so the identity stays in the audioId.

If the track is already the right song and only the runtime is short, extend. If the chorus is wrong, do not extend it — that writes more of the wrong chorus. Generate again, or cover into a new arrangement, then extend the one you would actually ship.

Do not chain five extends hoping the fifth will become a different genre. Style drift is not a feature you want on a brand bed. One continue from a chosen point. Listen. Stop.

FAQ

Can I extend a song I did not generate in Versely?

Not with this capability. extend_music needs the source track's audioId from a prior generate_music result. An upload is a different object. Cover can take a URL; extend cannot.

Is extend cheaper than generating a longer song from scratch?

It is a different purchase: more of the keeper versus a new lottery ticket. The credit figure is shown before you confirm either way. The waste is not the sticker. The waste is paying for a new identity when you meant more of the old one.

Where do I set the continue point?

continueAt is the optional field for "from the second chorus" or "near the bridge." If you omit it, the tool continues from or near a default point in the source. Name the moment when the moment matters.

Should I extend before I split stems?

Yes if you want stems of the long version. Split after the runtime is right. Extending after a stem split is not how this tool works — it wants the original Suno ids, not a vocal-only file you just made.