flutter_face_liveness_detection 0.2.8

On-device face liveness detection for Flutter with randomized active challenges and multi-signal passive anti-spoofing.

Changelog

0.2.8

• Fix: the instant-capture image (fastCapture) came out upside down / sideways. It now uses its own upright rotation per platform (Android raw sensor frames, iOS frames that arrive already upright and mirrored) instead of ML Kit's rotation value.
• Add LivenessConfig.fastCapture (default true): the final image is built from the latest camera frame where the face was straight (straightPoseMaxDeg, default 12), encoded on a background isolate. No preCaptureDelay pause, no shutter, no stream restart - capture takes milliseconds instead of ~1-2 s. Image is at preview resolution; set fastCapture: false for the previous full-resolution takePicture still.
• passiveFrameCount default lowered from 12 to 8 for a faster no-challenge session.
• The result is returned before the camera stream finishes stopping.
• One session at a time: add LivenessScreen.open(context, ...), which returns null without pushing if a screen is already open, and LivenessScreen.isOpen. A second LivenessScreen pushed while one is showing closes itself without opening the camera.
• Closing the screen (close button, back gesture, or result) always releases the camera, ML Kit detector and sensors. A new session waits for the previous one's camera to be released before opening it, and a pending start() completes as cancelled on dispose().
• Behaviour change: LivenessScreen no longer shows the "Verified" / "Verification failed" message or result card when a session ends - it just closes and returns the LivenessResult. Pass showResult: true (also on LivenessScreen.open) to bring them back; showCapturedImagePreview only applies with showResult on.
• Refreshed LivenessScreen design: top bar with round close button and title (messages.title); oval frame that is white while the face needs adjusting and green once it is positioned; a Face ID-style tick ring around the oval that lights up with progress (also in no-challenge sessions, via the new LivenessState.scanProgress); a scan line sweeping the face while a step runs; the instruction in a frosted-glass card with a per-step icon (spinner while starting or checking); step dots under the instruction; and a privacy footer (messages.privacyNote, empty hides it). "No face detected" now reads as the fix ("Position your face in the oval").
• The debug face box and landmark dots are off by default; pass showFaceOverlay: true to LivenessScreen to show them.
• Example app: starts with no challenges and mask detection off, and opens the screen through LivenessScreen.open.

0.2.6

• Challenges are now optional: pass challengeCount: 0 (or an empty challengePool) to skip active challenges entirely. The session collects face frames, then scores on the passive anti-spoof signals only (brightness, sharpness, glare, texture, micro-motion, mask check). Previously challengeCount was clamped to at least 1 and an empty pool fell back to a blink challenge.
• Add LivenessConfig.passiveFrameCount (default 12): frames to collect before scoring when no challenges are requested. Capped at analysisWindow.
• Example app: challenge picker now allows deselecting every challenge.

0.2.5

• Add LivenessConfig.messages (LivenessMessages): every user-facing string the package shows - status copy ("Position your face in the oval", "Hold still", "Checking", ...), per-challenge instructions (including holdStillInstruction, default 'Hold still and look at the camera'), result-card copy for each LivenessFailure, and the error panel/permission copy - is now overridable without forking the UI. Every field defaults to the current copy, so leaving messages unset changes nothing.
• LivenessConfig.instructionBuilder still takes priority over messages's per-challenge fields when both are supplied.

0.2.3

• Add requireFaceInOval (default false): when on, a detected face outside the on-screen capture oval is treated like "no face" - challenges won't advance and the final still won't be captured - until it's recentred.
• Captured still now crops to the face by default (cropToFace, default true) instead of saving the full camera frame, using the last known-good face box from the session plus a configurable margin (faceCropPadding, default 0.4). Set cropToFace: false for the old full-frame capture.
• Add LivenessScreen.showCapturedImagePreview (default false): shows a thumbnail of the captured still on the result card when the session passes.
• Add ovalCrop (default true): when cropToFace is on, the captured still is cropped to the on-screen oval's bounding box with everything outside the ellipse painted black, instead of a padded rectangle around the face box. Set ovalCrop: false for the old rectangular face crop.
• Oval size/position is now configurable: ovalWidthFraction (default 0.72, fraction of screen/frame width), ovalHeightRatio (default 1.32, oval height as a multiple of its width), and ovalCenterYFraction (default 0.4, vertical center as a fraction of height from the top). Drives the on-screen oval, the requireFaceInOval gate, and the ovalCrop capture shape consistently.
• Add captureCropScale (default 1.0): shrinks (or grows) just the ovalCrop capture area around the oval's center, independent of the on-screen guide oval - e.g. 0.8 crops the final still tighter on the face while the displayed oval stays the same size.

0.2.1

• Add preCaptureDelay (default 900ms): a "Hold still" pause between the last challenge passing and the final photo, so the still isn't snapped mid-turn right after a turn/nod challenge. Set to Duration.zero for the old instant-capture behaviour.

0.2.0

• Add mask detection: a worn mask flattens the nose/mouth region relative to the eyes/forehead, and the analyzer now watches for that texture drop every frame. A session that looks masked for too long, or too consistently, fails fast with LivenessFailure.maskDetected instead of completing challenges or capturing a photo. Tunable via enableMaskDetection, maskDetectionFrames, maskTextureRatio, maskEntropyDrop, maskSessionFraction.
• Face detector now defaults to ML Kit's FaceDetectorMode.fast for quicker per-frame detection on mid-range devices.

0.1.1

• Add screenshots to README.

0.1.0

• Initial release: on-device face liveness detection with randomized active challenges (blink, smile, turn left/right, nod down, hold still) and a multi-signal passive anti-spoof analyzer (brightness, sharpness, glare, texture, micro-motion, motion correlation, depth parallax).
• Optional pluggable AntiSpoofModel for a trained network signal.