Split stems after the generated track is the bed you will mix

separate_music_vocals splits a track you generated in-app into vocal and instrumental stems. You pass taskId and audioId from a prior generate_music or extend_music. Karaoke, a remix, an instrumental bed under a VO you already have — those are mix problems. Mix problems come after you know which song is on the locked cut.

Split a song into stems is the job. Billed per split, on top of the original generate. Arbitrary uploads are not valid here; those go to isolate_audio.

Stems are children of a parent you must keep

If the parent song is a maybe, the stems are two more files you will not mix. Teams split early because the instrumental "might be useful." Usefulness is measured against a picture duration and a VO. Without those, you have a folder.

Picture lock tells you whether you even need a split. If the keeper is MOS and you will attach the full mix, you did not need stems. If the keeper has a VO, you wanted the instrumental stem of the bed you chose, not of the first generate that sounded fine in headphones.

Do not split as a way to inspect whether Suno buried the vocal. Listen to the parent. Split when you have a job for both halves, or for the half you will actually attach.

Order

1. Lock picture. Know whether VO exists.
2. Generate (and extend if needed) the bed for that runtime.
3. Split if you need the instrumental under a VO, or the vocal for a karaoke pass.
4. Attach the stem you meant. Do not attach the parent and the stem.

The AI music generator is where the parent comes from. Stem split is not a generator. It is a slicer of a keeper song.

estimate_cost for the split. The expensive mistake is a second parent generate because you split the wrong audioId and cannot find the right one. Look up history first.

FAQ

Can I split a client's uploaded master?

Not with separate_music_vocals. Use isolate_audio on a URL. Different tool, same later-step rule: only the master you will ship.

Should I split before extending the track?

No. Extend the parent to the runtime, then split the file you will mix. Splitting then extending the parent leaves you with stems of the short version.

Why would I ever need the vocal stem for a brand video?

Karaoke-style captions, a remix, or a hook that uses the sung line. If you have none of those jobs, you wanted the instrumental or the full mix, not a split.

Does the split refund if I delete the stems?

No. Treat it like any other billed pass. Do it once, on the parent you will attach.