Find your celebrity look alike in one photo

Four steps, all of them local. First the browser looks for a face and picks the largest one if there are several. Next it rotates, scales and crops that face onto a fixed 112 by 112 grid, so that differences in camera angle stop mattering. Then a neural network turns the aligned face into 512 numbers. Finally those numbers are compared with the same numbers for every celebrity in the library, and the closest ones float to the top.

The comparison step is genuinely fast: an ordinary laptop compares your face against several thousand stored faces in a couple of milliseconds. The slow part is the first-time model download, about 58 megabytes, which your browser then caches.

Why the celebrity photos are middle-career shots

Age changes a face more than almost anything else, which is why a photo of someone at twenty and a photo of the same person at sixty can read as different people to a face model. Every reference photo in our library was therefore chosen from the middle stretch of that celebrity's career. It gives you a fairer comparison, and it avoids the uncomfortable outcome where the tool confidently matches a grown adult to a picture of a child.

What we can and cannot see

We cannot see your photo, your result or your match score. There is no endpoint that accepts them, so there is nothing to leak, subpoena or accidentally log. The only measurement the site takes is an anonymous count of page views and a handful of interaction counts such as "an upload was started", with no image, no score and no identifier attached.

The one exception is worth stating plainly: if you paste an image URL instead of uploading a file, the browser fetches that URL both directly and through an image proxy at the same time, taking whichever answers first, because most sites block direct requests. The address you typed therefore reaches that proxy on every image-URL submission, and is visible to it. Uploading a file avoids the proxy entirely.