Look at your Mac. It unlocks.
Open-source face unlock for macOS — free, private, and it lives in your notch. No subscription. No cloud. No account. Your face never leaves the machine.
You close the lid, you open it, the lock screen stares at you.
FaceMac recognises you through the built-in camera and types your saved password at your own lock screen. The notch lights up, scans, draws a checkmark, and you're in. About a second, all up, and the camera light goes off.
- Free, forever. One binary, GPL-3.0, no trial, no licence key.
- Private by construction. Frames are never written to disk. Faceprints and your password never leave the machine.
- The notch is the interface. It grows out of the notch, pulses a green face glyph while it looks at you, then collapses into a tick.
- Knows when it's not you. A stranger gets rejected in ~0.4 s with a red shake.
- The camera gets off. 5 seconds without a match and it stops.
- On-device ML. SFace via CoreML, 128-d embeddings, calibrated per face.
| Notch overlay | Real NSPanel above everything, including the lock screen (SkyLight) |
| Face ID style animation | Green scan glyph → morph into a Lottie checkmark |
| Guided enrolment | Five prompts — straight, left, right, tilt, tilt — with a live ring |
| Rejection | Confidently-not-you is caught instantly and shaken off in red |
| Liveness | Blink / micro-motion check rejects still photos, on-device |
| Calibration | The accept threshold is tuned to your face, camera and lighting |
| Anti-drift | Best-reference and centroid must agree, over N consecutive frames |
| Quality gates | Tiny or heavily turned faces are ignored instead of guessed |
| Retry button | A glass button on the lock screen to try again |
| Bilingual+ | English, Русский, 中文 — system auto-detect or manual |
| Privacy | Everything local, camera stops when it's done |
camera frame
→ Vision face detection + 5 landmarks
→ similarity alignment to the 112×112 ArcFace frame (FaceAligner)
→ SFace CoreML embedding, 128-d, L2-normalised (MLFaceEmbedder)
→ cosine similarity vs enrolment (FaceMatcher)
→ best + centroid + 3 consecutive frames must agree
→ blink / micro-motion check (LivenessTracker)
→ type the Keychain password with CGEvent (KeyboardInjector)
No frame is ever written to disk or sent anywhere.
Grab the latest release, drag FaceMac onto Applications, and on first launch right-click → Open (builds aren't notarized yet).
Not in the official homebrew/cask tap yet (releases aren't notarized), so it
ships as its own tap:
brew tap c1osed1/facemac https://github.com/c1osed1/FaceMac.git
brew install --cask --no-quarantine c1osed1/facemac/facemacThe cask lives at Casks/facemac.rb.
git clone https://github.com/<you>/FaceMac.git
cd FaceMac
scripts/fetch-model.sh # builds the SFace CoreML model (once, needs Python 3.12+)
scripts/install.sh # builds and installs to ~/Applications, then launches- Set Saved Password… — your macOS login password, into the Keychain.
- Grant Accessibility… — so it can type that password at the lock screen.
- Enroll Face… — five quick head poses.
- Keep Enabled, and lock your Mac (
⌃⌘Q).
Preview Notch Animation shows you the whole thing without locking.
- Apple silicon Mac, macOS 14 or later
- Built-in camera (Continuity Camera is not used)
Permissions are tied to a stable code signature, so a rebuild keeps the Camera/Accessibility grants.
FaceMac is a convenience, not a security upgrade.
- An on-device liveness check (blink / natural micro-motion) rejects a still photo, but it is not spoof-proof: a good video of you on a phone screen might still get in. Turn it off or relax it in Settings → Recognition if it gets in your way.
- It stores your macOS login password in the Keychain, because that's how it types it for you. It never leaves the machine — but it is on the machine.
- It cannot lock you out: if it fails or you quit it, you log in exactly as before.
- Recognition never runs while the screen is unlocked; the camera is off unless a scan is in progress.
swift build # core library + CLI
swift test # unit tests, incl. a CoreML vs ONNX parity test
xcodegen generate # produce FaceMac.xcodeprojLayout:
Sources/
FaceMacCore/ capture, Vision alignment, SFace embedder, matching,
Keychain, lock watcher, CGEvent input, coordinator
FaceMacApp/ menu-bar app, notch overlay, settings window
FaceMacDemo/ headless CLI (info / enroll / match)
Tools/Conversion/ SFace ONNX → CoreML conversion + golden-embedding generator
Tools/Icon/ app icon generator
scripts/ fetch-model.sh, install.sh
Releases are automated by GitHub Actions (.github/workflows/ci.yml):
- every push / PR builds the app and uploads a DMG artifact;
- a commit whose message contains
[RELEASE] 0.2.0, or a manual run of the Build workflow, publishes a GitHub Release with that DMG attached; - the version comes from
MARKETING_VERSIONinproject.ymlunless one is given explicitly.
Build it locally with:
xcodebuild -project FaceMac.xcodeproj -scheme FaceMac -configuration Release \
-derivedDataPath .build/ReleaseData build
scripts/make-dmg.sh .build/ReleaseData/Build/Products/Release/FaceMac.app FaceMac-0.1.1.dmgThe CI build is ad-hoc signed, so macOS warns on first launch. Warning-free
builds need a paid Apple Developer account: sign with a Developer ID
Application certificate, notarize with notarytool, and staple the ticket.
- On-device SFace embeddings + per-face calibration
- Notch overlay above the lock screen
- Guided enrolment, instant rejection, lock-screen retry button
- Liveness (blink / micro-motion)
- Homebrew cask
- Multiple faces per Mac
- Pre-login (FileVault) support via a privileged helper
- SFace — OpenCV Zoo, Apache-2.0. The recognition model.
- Atoll — GPL-3.0. The notch overlay approach.
- SkyLightWindow — MIT. Window above the lock screen.
- Lottie — Apache-2.0.
Not affiliated with Apple. "Face ID" is a trademark of Apple Inc.
GPL-3.0 — see LICENSE. Use it, fork it, ship it, just keep it open.