A face-recognition Swiss Army knife that runs in your browser tab

What it does

FaceAPI bundles pretrained TensorFlow/JS models for face detection, landmark tracking, face recognition, and demographic prediction (age, gender, emotion) into a single library. It works both in the browser and in Node.js, and ships with multiple build flavors: IIFE, ESM, and CommonJS, with or without TensorFlow pre-bundled.

The interesting bit

The original face-api.js was left unmaintained; this fork keeps the models alive while adding Node.js multiprocessing demos, GPU-accelerated builds, and a WASM backend for platforms where native TensorFlow binaries won’t run. That’s a lot of packaging work for what is essentially a thin wrapper around existing TFJS models.

Key highlights

• Pretrained weights included in the repo; no external model downloads at runtime
• Browser builds with TFJS baked in, plus a nobundle variant if you want to pin your own TensorFlow version
• Node.js support from v14 through v22, with explicit GPU and WASM backend paths
• Live webcam and static-image demos hosted on GitHub Pages
• Worker-pool example showing how to parallelize face detection across CPU cores

Caveats

• Node.js v23+ is explicitly unsupported due to TensorFlow/JS incompatibilities
• Using graphical functions in Node.js requires manually patching the environment with a canvas library
• The README is thorough but reads like a configuration manual; actual model accuracy and performance numbers are absent

Verdict

Worth a look if you need client-side face analysis without API keys or network round-trips. Skip it if you need production-grade accuracy benchmarks or a managed, automatically updated model pipeline.