Four-stem AI separation. Drop a track and walk away with clean stems for remixes, sampling and karaoke.
Up to 45 MB, any common format.
Four-stem split runs on Replicate GPUs in ~25 seconds.
Preview each stem, then download what you need.
Demucs v4 — the same neural model used by professional re-mixers.
High-bitrate MP3, full sample rate, ready for any DAW.
Clean output, no audible tags, no UI overlays.
Four: vocals (lead + ad-libs), drums (kit + percussion), bass, and music (everything else — pads, guitars, FX).
320 kbps MP3, full original sample rate. Lossless WAV is on the roadmap for Pro/Studio.
No. Clean output you can drop into any DAW or sampler.
About 25-35 seconds for a 4-minute track on Replicate GPU. Slower locally if Replicate is down.