mml_face_sdk 0.1.1-dev.3

Offline, on-device 1:1 face verification and passive liveness SDK.

MML Face SDK

Production-oriented Flutter plugin for offline, on-device 1:1 face comparison and passive liveness on Android and iOS. Native code performs face detection, five-landmark alignment, TensorFlow Lite inference, quality gating, and decision state. Model weights are not published with this package: activation installs a privately delivered, per-activation encrypted model pack into app-private storage. The service never receives images, video, embeddings, or templates.

Status: pre-release integration baseline. The bundled thresholds (0.70 similarity and 0.75 liveness) reproduce the source application behavior. They are not independently validated operating points. Calibrate against representative users, devices, capture conditions, and attack media before production.

Capabilities

• Enrollment/template creation from one encoded image.
• Recognition-only 1:1 comparison.
• Passive liveness plus 1:1 verification using two MiniFAS models and a four-frame median.
• Single-face, minimum-size, pose, landmark, finite-output, and template-version gates.
• Seven-day evaluation and lifetime Ed25519 licences bound to the customer app and each activated installation, then verified fully offline.
• A reusable customer key activates multiple installations of one Android package or iOS bundle identifier; trial time starts once per app and cannot be reset by reinstalling.

Try the Android demo

Download the signed demo APK. It activates its demo-only licence automatically and is restricted to the demo package identifier. SHA-256: f832c79d01aa0ba408e894f04d165315554f8719a674a67ce87b307078fc85ca.

For a seven-day trial in your own app, contact Mobile ML Labs on WhatsApp with your platform and exact Android package name or iOS bundle identifier.

Quick start

import 'package:mml_face_sdk/mml_face_sdk.dart';

final sdk = MmlFaceSdk(
  publicLicenseKey: SimplePublicKey(publicKeyBytes, type: KeyPairType.ed25519),
);
await sdk.initialize(license: tokenStoredInSecureStorage);
final template = await sdk.createTemplate(enrollmentJpegBytes);
final recognition = await sdk.recognize(encodedImage: probeJpegBytes, template: template);
final verified = await sdk.verify(encodedImage: nextLivenessFrame, template: template);

For liveness, submit four consecutive same-session frames/captures. Call resetLiveness() when the subject, camera session, or flow changes. The demo uses the system camera picker for portability; a shipping app should integrate a guided native frame stream at a controlled cadence.

See integration guide, API reference, threat model, attack statement, privacy notes, model inventory, and release checklist.

Development

flutter pub get
flutter analyze
flutter test
cd example
flutter run --dart-define=MML_LICENSE_PUBLIC_KEY=vJZyhTYJIS4JWop8tIlY1Juyzpyv434M_BUQhgHzgDY --dart-define=MML_ACTIVATION_URL=https://mml-face-license.hamzaasif19974-69b.workers.dev/v1/activate

The source attendance app is not a dependency and is not modified. This public package contains no biometric model weights. Commercial activation and a private model pack are required for inference.

Support: [e-mail]

Website: mobilemllabs.com · Source and issues