Reimagine & Remove vocals

Two features work from an audio file you upload instead of from a text prompt:

• Reimagine takes the lyrics and the length of your song and has Finetuning write and perform a brand-new track in a style you choose. It does not use the recording itself: no melody, no instruments, no voice carries over.
• Remove vocals takes the recording and returns it with the singing taken out — the same song, as an instrumental.

Both start with the same upload. Both finish as an ordinary generation that you poll with GET /v1/generations/:id or receive on a webhook, exactly like music you generate from a prompt.

How it works

1. POST /v1/songs/uploads          → upload id  (Reimagine: we start listening for the lyrics — 1 credit)
2. GET  /v1/songs/uploads/:id      → status "transcribed" + the lyrics we heard   (Reimagine only)
3. POST /v1/reimagine              → generation id      (1 more credit the first time, 2 after that)
   POST /v1/remove-vocals          → generation id      (1 credit)
4. GET  /v1/generations/:id        → audioUrl when status is "completed"  — or wait for your webhook

When you call POST /v1/reimagine without lyrics, we sing the transcript. Send your own lyrics to fix a misheard word or to write new ones — you don't have to wait for the transcript at all if you do.

Your rights come first

You may only upload a song you have the rights to use. Every upload needs an explicit confirmation — the API will not accept one without it.

Reimagine

I wrote this song, or I own or have licensed the rights to use its lyrics and composition.

I won't use this to copy another artist's song or voice, and I'm responsible for how I use what I make.

Remove vocals

I made this recording, or I own or have licensed the rights to use this recording.

The current statement version is 2026-09-30. GET /v1/songs/config always returns the current version and wording; read it from there rather than hard-coding it. When the wording changes, the version changes too, and an upload sent with an old version is refused with ATTESTATION_REQUIRED until your user has seen the new wording.

We keep a record of every confirmation: the feature, the statement version, the exact statements, a one-way hash of the caller's IP address and the User-Agent. It is kept after the upload itself is deleted.

Reimagining a song that was uploaded for Remove vocals needs the Reimagine statements too, because that confirmation only covered the recording. Send them with POST /v1/songs/uploads/:id/transcribe.

Limits

Credits and plans

So one Reimagine costs 2 credits, as it always has; three styles from one upload cost 1 + 1 + 2 + 2 = 6. The same applies to a track from your own Library, because we still listen to it.

• Transcription failed? Its credit comes back automatically. Your first retry with /transcribe is free; every retry after that costs 1 credit (and is refunded too if it fails). Transcribing again after a successful transcription also costs 1 credit.
• Reimagining a Remove vocals upload? Calling /transcribe on it costs 1 credit, which counts toward the first Reimagine the same way.
• Not enough credits to listen? POST /v1/songs/uploads for Reimagine answers 402 INSUFFICIENT_CREDITS before anything is uploaded or charged.
• GET /v1/songs/uploads/:id tells you what the next render will cost (nextReimagineCredits).

Credits come from your monthly allowance first, then from credit packs. If a job fails — at submission or on our GPUs — the credits go back to where they came from, automatically and exactly once. A failed first Reimagine hands the transcription credit back to the upload, so the next try costs 1 again.

Private, always

Tracks made with Reimagine or Remove vocals are built from someone's own song, so they stay private to your account:

• they can't be made public or shared with a link;
• they never appear in Discover, charts, search or public playlists, and can't be added to a public playlist;
• they don't get a generation certificate — GET /v1/generations/:id/certificate answers 403 PRIVATE_ONLY, because a certificate attests that Finetuning's model created the work.

You can download them and use them as your rights in the original song allow.

Webhooks

Pass webhook to POST /v1/reimagine or POST /v1/remove-vocals and we POST the finished (or failed) track to it, with the same payload as every other generation plus a type field ("reimagine" or "remove_vocals"). There is no webhook for the transcript: poll GET /v1/songs/uploads/:id, or skip the wait by sending your own lyrics.

Zapier's "New Completed Track" trigger doesn't fire for these tracks yet.

Errors

These are the errors specific to Reimagine and Remove vocals. Every error has the usual shape — see Errors.

Endpoints

• GET /v1/songs/config — prices, limits, upload caps and the current rights statements
• GET /v1/templates — styles for Reimagine
• POST /v1/songs/uploads — upload a song (file or URL); Reimagine uploads are transcribed for 1 credit
• GET /v1/songs/uploads — your uploads
• GET /v1/songs/uploads/:id — status, transcript and tracks made from it
• POST /v1/songs/uploads/:id/transcribe — listen again, or listen to a Remove vocals upload (1 credit; the first retry after a failure is free)
• POST /v1/reimagine — 1 credit after a paid transcription, otherwise 2
• POST /v1/remove-vocals — 1 credit