Can Audio Separation split any song, or only songs made in Nodaro?

Any audio. It works on an uploaded song, a recording, the sound of a video or a generated track. Suno Separate, by contrast, only accepts tracks made with Suno.

Which stems can I get?

In Vocal / Instrumental mode, a vocal track and an instrumental track. In Full stems mode, vocals, drums, bass and other, plus guitar and piano with the Auto quality.

Which quality should I choose?

Auto picks the best model for the mode and is the default. Fast is the quickest and cheapest. Best is a finer, fine-tuned model that is about four times slower and costs more.

How do I make a karaoke track?

Use Vocal / Instrumental mode and wire the Instrumental output onward. The Vocals output gives you the acapella.

Audio Separation node — split any song into vocals, instrumental or stems

Split any audio into vocals and instrumental, or into stems such as drums, bass, guitar and piano, with the Demucs model. Works on songs from any source.

The Audio Separation node splits any audio into its parts. In Vocal / Instrumental mode, it returns a clean vocal track and a music-only track. In Full stems mode, it returns vocals, drums, bass and other, and on the highest-quality model also guitar and piano. It runs Demucs, Meta's source-separation model, and accepts a song or a recording from any source.

When to use it

Take the instrumental backing track out of a finished song.
Pull an acapella vocal for a remix or for lip sync.
Split a track into stems to remix it.
Isolate the drums or the bass to sample them.

For a clean voice alone, with no instrumental track, use Voice Extractor.

Quick start

Add the node

Press Tab on the canvas and choose Audio › Clean & Separate › Audio Separation.

Wire the audio

Wire an audio node into the Audio input: an upload, a generated track or the sound of a video.

Choose the mode

Open the settings panel. Keep Mode on Vocal / Instrumental for two tracks, or choose Full stems.

Run it

Click Run on the node. Each part appears on its own output, and the settings panel has a player for the vocal and instrumental tracks.

A song is split, the vocals get a hall reverb, and the two parts are mixed back together.

Input and outputs

Input	Accepts	What it does
Audio	Audio nodes, such as Upload Audio, Reference Audio, Suno Create Music and Extract Audio	The audio to split.

Each part has its own output, and each output carries one audio file.

Mode	Outputs
Vocal / Instrumental	Vocals, Instrumental
Full stems	Vocals, Drums, Bass, Other, and Guitar and Piano on the Auto quality

Outputs that the chosen mode and quality do not produce stay inactive. The node's main result is the vocal track.

Settings

Setting	What it does
Mode	Vocal / Instrumental (the default) returns two tracks. Full stems returns one track per instrument group.
Quality	Auto (the default) picks the best model for the mode, including the six-stem model with guitar and piano in Full stems mode. Fast uses the base model. Best (slower) uses a fine-tuned model that is about four times slower.

Credits

The price is fixed per run, whatever the length of the audio. It depends on the quality, and Fast is the cheapest choice.

Tips

Karaoke and acapella. Use Vocal / Instrumental mode: the Instrumental output is the backing track, and the Vocals output is the acapella.
Guitar and piano. Use Full stems with Auto quality to get separate guitar and piano stems.
Songs from anywhere. This is the node for songs that were not made with Suno. Suno Separate refuses them.
Effects on one stem. Wire a single stem through Audio FX, then put the parts back together with Mix Audio.

From the API

POST /v1/audio-separation takes audioUrl, and the optional mode (vocal_instrumental or stems) and quality (auto, fast or best). The MCP tool is separate_audio. See Voice and media.

Audio Separation