Introduction

FrameFind is a collection of face analysis detectors — glasses detection, blink detection, head pose estimation, mask detection, and gaze estimation — built on top of MediaPipe and ONNX Runtime. It ships in two flavors: a React package with hooks that handle the video loop and model lifecycle for you, and a core package with bare classes for when you need more control.

All models load from FrameFind's CDN by default, so you can get started without hosting anything.

Packages

@framefind/react — React hooks for real-time webcam detection. Each hook manages the model, the video frame loop, and state updates internally. You attach a videoRef to your <video> element and read results from the returned state.

@framefind/core — Framework-agnostic detector classes. Use these in vanilla JS, in web workers, or in Node.js pipelines. You call load() once, then detectFromVideo() or detectFromImage() on each frame yourself.

Installation

# React (includes @framefind/core as a peer)
npm install @framefind/react

# Core only (browser or Node.js)
npm install @framefind/core

The React package re-exports everything from core, so you rarely need to install both.