AudioSeparator
Category: karaoke/separation. Runs a source-separation model on an incoming AUDIO and returns two stems as AUDIO. | Input | Type | Notes | |---|---|---| | audio | AUDIO | Standard ComfyUI audio dict | | modelfilename | dropdown | Model to run (pulled from audio-separator's manifest, falls back to a curated list) | | outputformat | wav \| flac | Intermediate file format used during separation | - primarystem (AUDIO) — first stem the model produces (e.g. Vocals for a vocal model) - secondarystem (AUDIO) — second stem (e.g. Instrumental) - model (STRING) — the selected modelfilename, passed through so downstream save nodes can compose filenames like vocals .wav Stem semantics depend on the model; the names are deliberately generic. Consult the model's documentation to know which stem is which. Models are downloaded on demand to ComfyUI/models/audioseparator/ and reused thereafter. Outside a ComfyUI environment (e.g. running the smoke test standalone), the default cache location is /.audio-separator/models/. If you already have a models folder mounted elsewhere (shared between UIs, on a separate drive, etc.), point ComfyUI at it via extramodelpaths.yaml: yaml my-setup: basepath: /mnt/models audioseparator: audioseparator/ The node searches every registered audioseparator folder for the selected model. If the file is already present in any of them, that folder is used directly (no re-download). If it isn't present anywhere, the first registered folder is used for the fresh download — entries in extramodelpaths.yaml load before custom nodes, so a configured external path takes p…
