A small tool with one strong opinion

Star By Face is a free browser tool that compares your photo with a library of celebrity faces and shows you the twelve closest. No account, no upload button in the usual sense, and no server holding your picture: everything happens inside the tab you already have open.

It is built by a small independent team. We made it because the alternatives all asked for something we did not want to give: an email address, an app install, or a photo uploaded to a machine we would never see.

Why we built it

The idea is not original. Dozens of versions exist, and most work the same way: you hand over a photo, a server runs the math, and you get a list back. That is a reasonable design, and an unnecessary one now. Face detection and face embedding were built to run on phones, and browsers have been able to run them directly through WebAssembly for years. If the whole pipeline fits inside a tab, why does anyone need your photo?

Photos never leave your device

Think of it as a calculator rather than a website. Your browser downloads the model files once, like a font or an image, and keeps them in its cache. Everything after that happens on your own hardware.

Your photo is read from disk into the tab's memory, drawn onto a canvas, and resized. A neural network finds the face and its landmarks. The face is straightened, cropped to a 112 by 112 pixel square, and converted into 128 numbers describing its structure. Those numbers are compared with the numbers for every celebrity in the library, and the closest twelve are ranked.

At no point does anything reach for the network. There is no upload endpoint, because there is no backend: the site is a folder of static files served by a content delivery network. We could not look at your photo if we wanted to. Close the tab or reload the page and the image is gone from memory. Open developer tools and watch the Network tab during a match to see for yourself.

Why it is free

Because it is cheap to run. No servers doing the matching, no GPU bill, no database, no storage. What remains is a domain name and the bandwidth to serve a few static files, which is a rounding error rather than a business.

That shapes what the site can be. Free tools usually pay for themselves with advertising, and ads on a face-matching page would mean third-party scripts watching what you do. That would contradict the one thing this site exists to prove. So there are no ad networks here and nothing to sell, because we collect nothing to sell.

How the celebrity library is built

Every celebrity photo comes from Wikimedia Commons under a free licence, either Creative Commons or public domain. Each image keeps a credit line naming its photographer or source and links back to its Commons file page. The full list is on the credits page, which is long on purpose.

One curation choice is worth explaining. Reference photos were picked from the middle period of each celebrity's career, roughly ages 30 to 55, rather than the most famous image of them at 22 or the most recent at 70. Face comparison works best when both photos show a face at a similar life stage, so mid-career portraits give most users a fairer comparison.

The honest consequence is coverage. Freely licensed portraits are not evenly spread across the world's famous faces, so our library leans Western and English-language. Some users will see results drawn from a narrower pool than they deserve, and we would rather say that plainly than pretend the library is universal.

What this site deliberately does not do

• No ad networks, no third-party trackers, no advertising identifiers.
• No accounts, no sign-in, no email collection.
• No stored photos. Nothing about your image or your result is kept.
• No selling or sharing of data, because there is no data to sell.
• No use of your photo to train anything.
• No claims that this is identification, biometric verification, or a security tool.

Known limitations

• This is entertainment. Face embeddings can rank resemblance, but the method has no scientific validation as a look-alike test.
• The match score out of 100 is a similarity score, not a probability, an accuracy rate, or a confidence level.
• One reference photo per celebrity means you are compared against a photograph, not a person, so results shift when your photo changes.
• The library skews Western and English-language, as described above.
• A modern browser with WebAssembly support is required, which excludes very old ones.
• There is no history or saved results, because saving them would mean storing them.

Who we are, and how to reach us

A small independent team, working on this alongside other things. No investors, no roadmap full of features we cannot support. We read every message, though replies can take a few days.

Found a bug, spotted a missing credit, or want to suggest someone for the library? Email hello@starbyface.net. Questions about how your data is handled are answered in technical detail on the privacy page, and the terms page covers the rest.

Try it

The tool is on the home page: one photo, twelve names, no trace left behind. To understand the process first, read how to find your celebrity look alike.