DEV Community: Bradley Harper

Why I built an AI face rater that runs fully in your browser

Bradley Harper — Tue, 22 Sep 2026 12:58:52 +0000

Why I built an AI face rater that runs fully in your browser

A few weeks ago I got curious: can a face analysis model run entirely in the browser? No server, no upload, no API key. Just drop in a photo and the model does everything locally.

That question turned into FaceGrade — an on-device AI face rating tool.

What it does

Scores facial symmetry and proportions
Runs 100% client-side (WebGPU / WASM)
Nothing leaves your machine
Free, no signup

Why on-device matters

Most AI face tools send your photo to a remote server. With on-device inference, the privacy story is honest: the image never leaves the tab. It also means no server costs, so the tool can stay free.

Try it: https://aifacerater.online

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

Bradley Harper — Tue, 22 Sep 2026 10:54:18 +0000

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.

I built an on-device AI face rater that runs 100% in the browser

Bradley Harper — Mon, 21 Sep 2026 13:14:58 +0000

Face-rating tools have always been a bad experience: upload a photo, wait for a server round-trip, sit behind a login wall, and get a watermarked single number. I wanted the opposite — a tool that measures facial geometry honestly, runs entirely on the device, and never asks for an account.

The result is FaceGrade, a free browser-based AI face rater. Here's how it works.

The model: 478 landmarks, on-device

Instead of calling a backend, the page loads MediaPipe's FaceLandmarker and runs it through WASM in the browser. One selfie produces 478 3D landmarks, and from those we derive the measurements people actually care about:

Facial symmetry — left vs. right landmark distances
Golden ratio — inter landmark proportions
Canthal tilt, jawline, midface and face shape
A 1–10 attractiveness score with a breakdown of what drives it

Nothing is uploaded. The image you pick is read locally and the landmarks are computed on your CPU/GPU, which is why the result appears in about a second and why there's no queue.

Why on-device matters here

Face photos are among the most sensitive images a person can upload. Every cloud-based rater has to ship that photo to a server, and most of them quietly train on it. Running the model locally means:

No account, no email, no watermark
No third-party tracking of your face
Works offline after the first load

The stack

It's a small TanStack Start app on Cloudflare Workers. The heavy part is purely client-side: MediaPipe tasks loaded from CDN, a custom score formula on top of the landmarks, and a downloadable report card so you can compare measurements over time.

If you've been using looksmaxxing-style tools that make you sign up, give it a try: FaceGrade — free AI face rater. Upload one photo and you'll get the full measurement breakdown in about a second.