Separate Stems (demucs)

Decompose an audio file into isolated instrument stems using Meta's demucs model. Produces WAV files for vocals, drums, bass, and other instruments — or 6-stem breakdown with the htdemucs_6s model.

When to use

• The user wants to isolate vocals or drums from a mixed recording.
• Preparing a song for karaoke or remixing.
• Extracting a vocal layer from a podcast with music bed.
• Building separated stems for music production or remastering.

Do not use this skill for:

• Quick vocal isolation with minimal processing time → use extract-vocals (faster 2-stem mode).
• Removing vocals entirely → use a different skill or post-process stems.

Inputs