MediaPipe FaceLandmarker

A high-performance machine learning solution that detects 478 3D face landmarks and blendshape scores in real-time across mobile and web.

MediaPipe FaceLandmarker delivers precise facial geometry estimation using a single camera feed. It maps 478 3D landmarks (including iris tracking) and outputs 52 unique blendshape scores to drive realistic digital avatars or AR effects. The system runs locally on-device via GPU or CPU (optimized for Android, iOS, and JavaScript) to ensure low-latency performance without cloud dependency. Developers use it to power virtual try-on tools, expression-based UI controls, and live motion capture for 3D engines like Unity.

https://developers.google.com/mediapipe/solutions/vision/face_landmarker