Separating vocals from a generated song needs that track's Suno IDs
separate_music_vocals splits a Versely Suno generation into stems. Any mixed URL is isolate_audio, not this job.
Separating vocals from a generated song needs that track's Suno IDs. One generated track. Two clean stems. separate_music_vocals splits a prior Suno generation into vocal and instrumental using that track's taskId and audioId. It will not split an MP3 you recorded in a car. Pointing this job at any mixed URL wastes credits on a call that cannot see the IDs it requires — or it quietly becomes a different isolation tool that is not a stem split of your Suno master.
Separate vocals from a generated song is the named capability. Reference the track you generated earlier. The agent looks up the ids from history. You get two files: vocal stem, instrumental stem. Priced per generation, shown before you confirm.
What this job actually is
You made a song with generate a song from a prompt (or an extend of that track). You now want the bed without the singer, or the singer without the bed. This is that split, for that catalog of ids.
It is not mastering. It is not a cover. It is not "remove the music from this voice memo." Those are other rows.
What wastes credits
Any mixed recording. A live band, a voice note, a YouTube rip, a voiceover with a bed you attached in the editor — that is isolate vocals from an audio clip (isolate_audio). Same English sentence, different required params. Stem split wants Suno ids. Isolation wants an audio URL. Mixing them is how you either error or get a vocal-ish file that is not the instrumental you needed for the recut.
You wanted a cover in a new style. Create an AI cover of a song reinterprets a URL. Stems keep the same performance, split. A cover is a new performance.
You wanted a new instrumental from a prompt. Generate an instrumental track. Do not stem-split a vocal song hoping the leftover bed is the cue you would have written.
Splitting a track you have not generated in Versely. There are no taskId / audioId to look up. Upload-and-isolate is the other job. Pretending a cover URL is a Suno generation does not mint the ids.
The test
Was this file born as a Versely Suno generation you can point at in history?
If yes, and you want vocal and instrumental as two files, this is the job.
If it is any other mix, isolate. If you want a new arrangement, cover. If you want a new song, generate. This is one named agent job. Using it for a different job wastes credits.
FAQ
Can I split a song I uploaded from my phone?
Not with this tool. This path needs the Suno taskId and audioId from a prior generate_music or extend_music result. For an arbitrary mix, use isolate vocals from an audio clip.
Do I have to copy the ids out of a log?
No. Reference "that last song" or the track from earlier. The agent looks them up. You still have to be pointing at a Suno generation, not at a random URL.
Is the instrumental stem the same as generating an instrumental cover?
No. The stem is the accompaniment from this take. A cover is a new take. If the original arrangement is the asset, split. If you hate the arrangement, cover or regenerate.
Why have two vocal-removal jobs?
Because the inputs are different and the outputs are different. One is a Suno stem split with two files. The other is a vocal isolation from mixed audio. Collapsing them in chat is how you spend a generation on the one that cannot see your file.