DEV Community

Bradley Harper
Bradley Harper

Posted on

Running an on-device AI face rater: what I learned

In my previous post I showed a face rater that runs 100% in the browser. A few follow-up notes after shipping it.

Why on-device

Most face-analysis tools upload your photo to a server. That is bad for privacy and slow. I wanted something where the image never leaves the tab.

The stack

  • A small vision model quantized to run in WebAssembly
  • Browser webcam / file input, no upload
  • Local post-processing for facial landmarks

Result

The score is instant, no account, no watermark. You can try it here: FaceGrade

Takeaway

For sensitive inputs like faces, local-first is the obvious choice: lower latency, no data transfer, nothing to store.

Top comments (0)