How it works.
For two decades, this site has covered the people who define popular culture. Now we’re handing the lens back to you — upload a photo and find out which celebrity you most resemble, with the same machine-vision pipeline that powers state-of-the-art face recognition.
The flow is short. You drop a photo, you tick two consents (one to confirm you’re an adult, one to confirm the photo is yours), and an InsightFace Buffalo_L model encodes your face into a 512-dimensional vector. That vector gets compared, by cosine similarity, against a curated library of fifty celebrities stored in pgvector. The closest five are returned, ranked.
Compute happens server-side in Falkenstein, Germany. The face service runs on CPU and typically returns a match in under 200 ms; the rest of the “two seconds” you see in marketing copy is upload time and result rendering. EXIF metadata is stripped before the photo ever touches the model — your camera details and location never leave your machine in a form we record.
The result page reports your top match plus four runners-up, with a face-shape classification and per-feature breakdown scores for eyebrow and smile. A shareable card is generated server-side, sized for both Twitter and Instagram. Then the source photo is deleted, unless you explicitly opted in to the public look-alike gallery.
“The closest thing to a verdict you’ll get from a computer.”
The science.
A face-recognition model isn’t looking at your face the way you do. It doesn’t know what eyes are. It learned, from millions of training images, which arrangements of pixels tend to belong to the same person — and produced an embedding function that compresses any face crop into 512 numbers such that two photos of the same person produce nearby vectors and two photos of different people produce far-apart ones. We use cosine similarity, which is essentially the angle between two vectors, to rank distance. Identical faces sit at angle zero; strangers are perpendicular.
This is recognition, not classification. The model isn’t picking from a closed list of celebrities — it’s placing your face somewhere in the 512-dimensional space and reporting the celebrities that landed nearest. That’s why the system can degrade gracefully on faces it has never seen before, and why adding new celebrities to the library is just a database insert. The face-shape, eyebrow, and smile scores come from a separate analysis of the 106-point landmark map — ratios of jaw width to face length, brow arch height, and the like.
Privacy as a feature.
Photos uploaded to the tool are processed in seconds and then deleted, full stop, unless you tick the “publish to the gallery” toggle on the result page. There are no accounts and no fingerprinting. EXIF metadata is stripped on upload, the face vector is never sold or shared with any third party, and UK and EEA visitors are blocked from upload until our DPA and DPIA are in place — we will not process biometric data without the right legal foundation.