Vision / Runs in your browser

Compare two faces in your browser: are they the same person?

Two photos, 68 landmarks each, a face-embedding distance and the proportions that stay put from childhood to adulthood.

  • Free, no account
  • Can use your camera
  • Fetches an open-source model or engine when needed

Photos are analysed inside this tab and never uploaded. Skin Scan and Face Match download an open-source face model (a few MB) from a CDN the first time.

Loading Face Match…

How it works

  1. 01

    Add two photos, one in each slot. Front-facing, eyes visible and no heavy filters give the steadiest result.

  2. 02

    Press Compare faces. The first time, an open-source face model (detector, landmarks and recognition net, about 7 MB) downloads from a CDN, then both faces are found in this tab.

  3. 03

    Each face gets 68 landmarks and a 128-value embedding. The embedding distance, the eye-spacing-normalised proportions and the vertical structure are compared.

  4. 04

    Read the match percentage and verdict, the proportion bars and the per-photo lighting, tilt and turn checks, then copy the summary if you want to keep it.

Questions

Are the photos uploaded?

No. Both photos stay in memory in this tab; detection and the comparison run in your browser. Nothing is stored and nothing is sent anywhere.

What is downloaded the first time?

The open-source face model: a detector, the 68-point landmark net and the recognition net, about 7 MB in total, fetched from the jsDelivr CDN and cached by the browser. If an ad blocker or privacy extension blocks jsdelivr.net, the model download stops and the tool shows an error with a retry.

Can Face Match prove two people are the same?

No. It is a similarity estimate from two photos, not an identification. Under 0.6 embedding distance usually means the same person and under 0.4 is a strong match, but lighting, age, weight, expression, glasses and filters all move the numbers. Never use it to make decisions about a person.

What makes a good pair of photos?

Faces at least 80 px wide, looking at the camera with the head level, evenly lit and in focus. The tool warns when a face is under- or over-exposed, soft, turned away or tilted, and always compares the largest face when several are in a photo.

Built with @vladmandic/face-api (MIT).