Support

Apple Silicon and macOS 14.2 or later. Intel Macs are not supported — the separation model needs more than they have.

Anything that makes sound: Apple Music, Spotify, a browser tab, a video player. karanow reads the system output, not any app’s API, so nothing needs to support it. Spotify and Apple Music additionally report the track and accept next, previous and pause.

Neither. Audio is processed on your Mac inside a rolling window in memory; when the window moves on, it is gone. It is never written to disk and never sent anywhere. The one thing written to disk is a recording you start yourself, which is your own voice, and it stays on your Mac.

No. The backing track exists while it is playing and is never saved. If you need a file, an offline separation tool is the right thing to use.

It has a hard budget: every 1.125 seconds it must separate 1.125 seconds of audio. On an M2 Max that takes 268 ms — about a quarter of the budget. The app shows the percentage your Mac is using, so you can see the headroom rather than guess at it.

It shows a source only when an app genuinely owns a live output stream. Start playback somewhere and the name appears in the top right. If every source stops, it goes back to waiting rather than leaving a stale app name up.

Two, once each. System audio recording lets karanow hear the app you are playing from. Microphone access is used only to measure the pitch of your singing. If you declined either, macOS System Settings → Privacy & Security is where you grant it.

karanow needs the AI model and a 44.1 kHz output rate. If either is unavailable it reports the reason instead of falling back to old-fashioned mid/side cancellation. If your output device’s sample rate changes mid-song, it passes audio through untouched and tells you why.

karanow can only read what a player publishes. Apps with an AppleScript dictionary answer and get lyrics, pitch and transport control; apps without one — Bilibili and WeSing for instance — still get real-time accompaniment. After three unanswered attempts karanow stops asking that app.

Candidates are checked against title, artist, album, duration and version tags to keep Live takes, remixes and covers out. If one still slips through, tell us the song and we will look at the match.

Then karanow will not show any. It can only display lyrics one of its sources already has, and when nothing matches you still get the backing track and the pitch line — just no words on screen.

That is deliberate and there is no switch. A Bluetooth headset can do high-quality playback or two-way calling, not both — claiming its mic drags your backing track into a 16 kHz mono call codec. With headphones on there is no speaker bleed to cancel anyway, so the Mac mic is the better input.

Also deliberate. Those devices only hear the computer’s own output, so they would draw a pitch line that matches perfectly whether or not you sing. If a Mac has nothing but these, karanow reports no usable microphone rather than making do.

Auto mode takes over a nearby iPhone wirelessly and switches straight back to the Mac microphone when it leaves, writing the reason on the runway rather than switching silently. You can pin the choice to Mac or iPhone on the vocal mix card.

The first 7 days are free, with no account and no card. After that it is £12.99 a year, shown in your local currency at checkout. In mainland China it is a one-year pass for ¥98, paid once by Alipay or WeChat Pay, with no renewal.

Settings → Account & subscription opens the Stripe billing portal. In mainland China there is nothing to cancel: the year is paid once and simply ends.

Write to us. A full refund revokes access; a partial refund does not.

Sign in with Apple or Google on the new Mac. Entitlements live on the server against your hashed identifier, so they follow the Apple account rather than the machine.