Audio Separation

Separate any audio into vocals + instrumental, or full stems (drums, bass, other, guitar, piano), using Demucs.

Overview

The Audio Separation node (internally audio-separation) uses Demucs (Meta’s Hybrid-Transformer source-separation model) on Replicate to split any audio into its component stems. Unlike Suno Separate — which only works on Suno-generated tracks — Audio Separation accepts any uploaded or upstream audio (a song, a recording, an extracted video track, etc.).

It has two modes:

Configuration

Field Type Default Description
Label string "Audio Separation" Display name for the node on the canvas
Mode "vocal_instrumental" \| "stems" "vocal_instrumental" Two-stem voice/music split, or full per-instrument stems
Quality "auto" \| "fast" \| "best" "auto" fast = base Demucs (htdemucs); best = fine-tuned (htdemucs_ft, ~4× slower); auto picks the best model for the mode (htdemucs_6s for full stems)

Inputs & Outputs

Outputs not produced by the chosen mode/model are inactive. The primary audio output defaults to the vocal track.

Credits

Fixed per run (reserved tier, not metered):

Quality Credits
Auto / Fast 3
Best 8

Best Practices

Common Use Cases