STEM — Live Stem Separation

Plug in your audio interface, start the record, and the vocal is on its own fader. STEM splits what is playing — or what is coming in from your mixer — into vocals, drums, bass and everything else, live, on the iPad in front of you. Each stem lands on its own pair of hardware outputs, so your desk does the mixing.\n\nNo upload. No render queue. No waiting for a track to be analysed before you can play it.\n\nSTEMS WHILE IT PLAYS\nLoad a track and it comes apart as it plays. Seek anywhere and the stems follow, because the separation is happening now rather than being read from a file somebody prepared earlier.\n\nLIVE IN, STEMS OUT\nFeed STEM from your interface — a turntable through your mixer, a booth output, an instrument — and it separates what is coming in, with the engine aimed at a live source instead of a file. The mix you did not make, taken apart.\n\nEVERY STEM ITS OWN OUTPUTS\nEach stem takes its own pair of hardware channels — vocals to 3-4, drums to 5-6 — up to 16 channels, assigned in a routing matrix that remembers each interface. Outputs start muted until you confirm the routing, because a misrouted stem into a PA is the mistake worth preventing. On a device alone the four stems fold down to stereo, and the console says so.\n\nTWO DECKS, ONE CROSSFADER\nOne setting runs a second deck: its own separation engine, its own four stems, its own routing. A crossfader between them, constant-power or cut.\n\nHANDS ON THE HARDWARE\nMap any MIDI CC by tapping a control and moving the knob. Soft takeover means a hardware fader never makes the mix jump, and positions are sent back, so motorised faders and LED rings follow the console.\n\nA CONSOLE, NOT A DIALOG\nLong-throw faders with a 0 dB detent, mute and solo per stem, LED metering, a master strip with BS.1770 loudness. Per-stem three-band EQ and a resonant filter — ladder or state-variable, your choice. Cue any stem to a headphone pair before you bring it up. Save the whole desk as a scene and recall it.\n\nSENDS, INSIDE OR OUT\nUp to four stereo sends, each per stem and post-fader. Send one to a hardware pair for your outboard, or to the built-in tempo-synced delay or plate reverb.\n\nTAKE THE STEMS WITH YOU\nWrite a loaded track back out as four 24-bit WAV files, latency-compensated so they sit at zero in a DAW, rendered by the larger offline model.\n\nMEASURED, NOT QUOTED\nLatency, load and real-time margin are read from the running session and shown while you work. The app never displays a number it has not measured.\n\nRUNS ON iPad AND Mac\nThe same app runs on Apple-silicon Macs. Real-time separation needs Apple silicon; older devices cannot run the model fast enough to keep up with playback.\n\nPRIVACY\nNo accounts, no analytics, no network. Your music never leaves the device.