Celebrity Look-alike
Upload a photo to discover which celebrity shares the most similar facial geometry. AI face matching runs entirely in your browser — for entertainment only.
Upload a clear, front-facing photo
Drag & drop or click to browse
Your photo stays on your device — it is never uploaded to any server.
How celebrity face matching works
Three-stage neural network pipeline, entirely client-side.
Face detection
TinyFaceDetector locates your face in the photo and extracts the face region with bounding-box precision. Works on any image size or aspect ratio.
Landmark alignment
FaceLandmark68Net maps 68 precise facial points and aligns the extracted face region, normalising for head tilt and scale before embedding is computed.
128-d embedding match
FaceRecognitionNet converts your aligned face into a 128-number vector. The closest matching celebrity embedding is returned as your look-alike.
What makes two faces similar?
The AI measures geometric similarity — not appearance, ethnicity, or skin tone.
128-dimension embedding
FaceRecognitionNet converts a cropped face into a 128-number vector. Each number encodes a specific geometric pattern — relative positions of eye corners, nose width, lip border, and dozens more facial measurements.
Euclidean distance
Two faces are compared by computing the straight-line distance between their 128-d vectors. Smaller distance = more similar facial geometry. A threshold filters out non-matches below confidence level.
Trained on millions of face pairs
The model was trained to minimise distance between photos of the same person and maximise it between different people, regardless of lighting, expression, or angle variation.
Not pixel matching
The comparison is entirely geometric — it does not compare pixel colours, skin tones, or photo backgrounds. Two very different-looking photos of the same person produce very similar embeddings.
100% private — no upload
All three neural network models — TinyFaceDetector, FaceLandmark68Net, and FaceRecognitionNet — run locally via JavaScript and WebAssembly in your browser. Your photo never leaves your device and disappears when you refresh or close the page.
Tips for the best match
- Front-facing photo — a straight-on angle produces the most accurate embedding
- Neutral expression — smiling distorts the geometry the model measures
- Good, even lighting on both sides of your face
- No sunglasses or items obscuring the eyes or jaw
- High-resolution photo — blurry images degrade the face detection step
- Hair pulled back to expose the full facial outline if possible
Celebrity Look-alike — frequently asked questions
How does the celebrity face matching work?
The tool runs FaceRecognitionNet (via face-api.js) in your browser to convert your photo into a 128-dimension numerical representation of your facial geometry. It then computes the Euclidean distance between your embedding and every entry in the celebrity database, and returns the closest match.
Does the tool actually recognise celebrities from my photo?
No — it measures geometric similarity between facial structures. A high match percentage means your face geometry (spacing, proportions, feature positions) is numerically close to a celebrity's encoding. It is not facial recognition in the identity-verification sense.
Why might the match percentage be low?
Match percentages are affected by photo quality, angle, lighting, and expression. The celebrity database also matters — if your look-alike isn't in the dataset, the best available match will still be returned but with a lower similarity score.
Is my photo stored or sent to a server?
No. Face detection, landmark mapping, and embedding generation all run locally using WebAssembly in your browser. Your photo is never transmitted and disappears from memory as soon as you close or refresh the page.
Why is the same celebrity appearing for every photo I try?
This can happen when the celebrity database is small. The tool returns the geometrically closest match from the available dataset. A larger, licensed dataset would produce more varied and accurate results across different users.
What photo gives the best match result?
A clear, front-facing photo with a neutral expression and good lighting produces the most accurate face embedding. Avoid extreme angles, heavy filters, sunglasses, or very low-resolution images — all of these distort the embedding and reduce match accuracy.