what is stems.

insert stems on a track in logic pro — or any audio unit host. press play and it listens to your track. or load a file, or drop one on the window.

pick a part: drums, bass, keys, pad or lead. one tap sets the session’s style, so every part sounds like the same band. add a few words about how it plays and sounds — “dusty breakbeat with ghost notes” — and press write.

the part comes back in whole bars at the host tempo, cut so it starts on the downbeat. feel off writes from your words alone; turn it up and the part starts from your track itself, so it sits on your groove. one press writes up to three takes; flip between them, re-roll, and drag the one that sits onto a track as plain audio.

• no vocals. the model can’t sing. it makes vocal-like textures, never words, so there’s no voice part.
• a key hint, not a lock. stems sends your track’s key; the part lands on it or a fifth away. re-roll if it clashes.
• not text-to-song. it writes one part at a time, for your track. the song stays yours.
• not a daw. your daw arranges. stems hands it audio and gets out of the way.

stems runs stable audio 3 medium — stability ai’s open model, in its fast 8-step form — on apple’s mlx, through a small background brain next to logic. an 8-bar part takes about five seconds on an m3 pro; one that follows your track, about seven.

the stems app sets the model up once: about 7 gb, fetched with your free stems account, or from hugging face without one. if stems loop is already installed, it uses that copy. after that, generation never leaves the mac — no upload, no queue, nothing to train on.

powered by stability ai. stable audio 3 is under the stability ai community license.

• apple silicon — m1 or newer.
• macos 14 or newer.
• about 8 gb free, for the model.
• an audio unit host. it’s made in logic pro; any host that loads au plug-ins can load it.
• a free stems account, if you want the model from the stems cdn. it’s the same key stems loop uses.