Suno Separate

The Suno Separate node splits any audio into its parts with Suno. Vocal / Instrumental returns a vocal track and an instrumental track. 12 Stems returns every stem Suno finds, such as drums, bass, guitar and piano. One instrument returns the instrument you name, alone, and the song without it. The audio can be a Suno track, an upload, another node's audio or a link.

When to use it

• Take the vocals out of a song to remix or sample them.
• Keep the instrumental for karaoke or as background music.
• Pull one instrument out of a mix, or remove it so that you can play or sing that part yourself.
• Split a song into stems to remix or master it.
• Prepare the vocals for a voice change or a dub.

Quick start

Input and outputs

Which outputs are active depends on the mode. Isolated and Without work only in One instrument mode, and Vocals and Instrumental are inactive there. An output with nothing behind it stays empty. It is never filled with the original audio.

• 12 Stems. Every stem appears in the settings panel with its own player and a download. The stems are not outputs that you can wire to other nodes.
• One instrument. Suno returns two takes. Take 1 drives the Isolated and Without outputs. Take 2 appears in the settings panel, where you can play and download it, but you cannot wire it.

Where the audio comes from

The node chooses the source by itself. There is no switch to set.

• A Suno song. When a Suno node is wired in, or a Task ID is typed, Suno separates its own track by its IDs.
• Any other audio. An upload, another node's audio or a link is sent to Suno as a file, up to 20 MB.

When the node has a Suno Task ID and other audio as well, it uses the Task ID and ignores the other audio. Clear Task ID to split the wired audio instead.

Settings

When Suno refuses

Suno does not separate a recording that it recognizes as commercially released. Audio sent as a file must also be 20 MB or smaller.

• If Suno refuses, use Demucs off (the default). The run fails with a clear message, and the credits are refunded in full. A file over 20 MB fails the same way, before anything is sent to Suno. Use Audio Separation for that audio, or turn the setting on.
• If Suno refuses, use Demucs on. The audio is separated with Demucs, the model behind Audio Separation, and the node shows a note that Demucs was used. A file over 20 MB goes straight to Demucs. Demucs returns its own stems: vocals and instrumental, or vocals, drums, bass, guitar, piano and other.
• One instrument never falls back. Demucs cannot pull out a named instrument. A refused One instrument run fails and is refunded in full, whatever the setting.
• A Suno song never falls back. Suno does not refuse its own tracks, and Demucs needs an audio file, so a run that separates a Suno track by its IDs does not use Demucs.

If Suno cannot download the audio at all, the run fails with "Suno couldn't download this audio" and is refunded in full. This is not a refusal, so it does not fall back to Demucs.

Credits

The price is fixed per run. It is the same for a Suno track and for any other audio.

A failed or refused run is refunded in full. When the Demucs fallback runs, you pay the Audio Separation price instead: 33 credits for vocals and instrumental, or 66 credits for full stems. That price is never more than the Suno price you were quoted. The Suno price is reserved when the run starts, and the difference is released when Demucs finishes.

Tips

• Start with Vocal / Instrumental. Move to 12 Stems only when you need each instrument separately.
• Listen to both takes. In One instrument mode, take 2 can be better than take 1. Download it from the settings panel.
• Keep files under 20 MB. Trim long or high-bitrate files first, or let the Demucs fallback handle them.
• Re-voice the singer. Wire the vocals into Voice Changer or Dubbing.
• Write new vocals. Wire Instrumental or Without into Suno Add Vocals to sing new vocals over the backing track.
• Commercial music. For a commercially released song, use Audio Separation, or turn on If Suno refuses, use Demucs so that a refusal does not stop the workflow.

From the API

The MCP tool is suno_separate_stems. Pass one of audio_url or audio_asset_id, and set type to separate_vocal, split_stem or split_stem_advanced with a stem_name. Set fallback_to_demucs to use Demucs when Suno refuses. See the MCP tools.

A finished job's output_data has one URL for each result: vocalUrl and instrumentalUrl, one field per stem in 12 Stems mode (for example drumsUrl, bassUrl and pianoUrl), and isolatedUrl, withoutUrl, isolated2Url and without2Url in One instrument mode. audioUrl is the main result. After a Demucs fallback, engineUsed is "demucs" and sunoRefused is true.