face_guard_liveness 1.0.4

A production-ready Flutter SDK for real-time on-device face liveness detection and face verification using ML Kit and TFLite.

Face Guard Liveness SDK

A production-ready Flutter package for real-time AI-powered face liveness detection and face verification.

Features

✅ Real-time face detection using Google ML Kit. ✅ Guided Liveness Actions: Blink, Smile, Turn Head, Look Up/Down. ✅ Face Verification: On-device face embedding comparison. ✅ Anti-Spoofing: Basic texture and motion analysis to prevent replay attacks. ✅ Fully On-Device: No data leaves the device. ✅ Clean Architecture: Easy to extend and maintain.

Installation

Add the following to your pubspec.yaml:

dependencies:
  face_guard_liveness: ^1.0.0

Android Setup

1. Change the minimum Android sdk version to 21 (or higher) in your android/app/build.gradle file.
2. Add camera permission to AndroidManifest.xml:

<uses-permission android:name="android.permission.CAMERA" />

iOS Setup

1. Add camera permission to Info.plist:

<key>NSCameraUsageDescription</key>
<string>This app needs camera access for face liveness detection.</string>

Usage

Initialize

await FaceGuardLiveness.initialize();

Start Liveness Check

final result = await FaceGuardLiveness.startLivenessCheck(
  context,
  actions: [
    LivenessAction.blink,
    LivenessAction.smile,
    LivenessAction.turnLeft,
  ],
);

if (result != null && result.isLive) {
  print("Liveness check passed!");
}

Face Verification

final result = FaceVerifier.verify(embedding1, embedding2);
if (result.matched) {
  print("Faces matched with similarity: ${result.similarity}");
}

TensorFlow Lite Models

This package requires TFLite models for face embeddings. You should place your .tflite models in the assets/models/ directory and list them in your pubspec.yaml.

Example:

flutter:
  assets:
    - assets/models/mobile_facenet.tflite

Security & Privacy

• All processing is done offline.
• No images are uploaded to any server.
• Biometric data is processed in memory and not persisted unless explicitly saved by the developer.

License

MIT