Find your celebrity lookalike with AI. In seconds.
Upload a photo and Ollie's machine learning model finds the five celebrities whose faces are closest to yours. Free, and your photo is never stored.
From photo to lookalike
Three steps, a few seconds. Your first search each day is free, no account needed.
- 01
Upload your photo
One clear, front-facing face works best. Your photo is only used for the search, then discarded.
- 02
Ollie reads your face
A detector finds and lines up your face. Then Ollie's neural network turns it into 512 numbers that describe its structure.
- 03
See your top five
Those numbers are compared with every photo of 5,000+ celebrities, and the five closest faces are ranked by similarity.
New: Looksmax AI
Rate your face out of 100. Then raise it.
Upload a selfie and Ollie scores your eyes, eyebrows, nose, lips, jawline, face shape, symmetry and hair out of 100, shows your potential score, and gives you a glow-up plan: the haircut, brows and beard that suit your face, plus skin and fitness targets. Free to try, and it runs on your device.
Rate my face freeNew: Ollie Stylist
Find the haircut that suits your face
Ollie Stylist is a free AI stylist. A 10-second face scan in your browser works out which of 7 face shapes you have and ranks the haircuts, beards and glasses that suit it. Then dress a realistic model in 81 real pieces from 37 brands, or get a full outfit in one of 9 styles, with a shop link for every piece.
Try the AI stylist free
More free face tools
How symmetrical is your face? Maps 478 points in your browser and ranks you against 3,800+ real faces, overall and for the eyes, brows, nose, mouth and jaw. Free to try, photo never uploaded.
Test my symmetryAre they the same person? Upload two photos and Ollie's face-recognition model (99.5% on the standard LFW test) shows how alike they are in seconds.
Compare facesOur own face-recognition model
Written and trained from scratch in PyTorch on a public research dataset of 5.8 million face photos of 85,742 people, instead of calling a big-tech AI service.
How Ollie works, in detailAccuracy on LFW
A benchmark that tests whether Ollie knows if two photos show the same person
Private by default
Your photo is held in memory for the search, then discarded. Never saved or used for training.
Trained from scratch
Trained on a public research dataset of 5.8 million face photos of 85,742 people.
Ollie's own network
A 20-layer convolutional neural network (SphereFace architecture) trained from scratch with CosFace loss.
5,000+ celebrities, all credited
Well-known living adults, 40,000+ photos in all (2 to 12 each), freely licensed from Wikimedia Commons and credited next to every match.
What Ollie is, and how the matching works
Ollie is a small, independent face-recognition project. Most lookalike apps send your photo to a general-purpose AI and ask it for a guess. Ollie doesn't. It runs its own model, a 20-layer convolutional neural network (SphereFace architecture) trained from scratch with CosFace loss, trained on a public research dataset of 5.8 million face photos of 85,742 people in about 10 days on one RTX 4060 Ti. On Labeled Faces in the Wild, the standard test of whether two photos show the same person, it scores 99.5%.
The celebrity side is a database of 40,000+ openly licensed photos of 5,000+ famous people from Wikimedia Commons, with 2 to 12 photos per person so that one flattering shot can't decide a match. When you upload a selfie, the model turns your face into 512 numbers that describe its structure and compares them with every photo in the database. Hair colour, makeup and the background barely move the score, because the model was trained to ignore them.
The percentage you see is a similarity score, not a probability. Two random celebrities score about 33%, and only one pair in a hundred reaches 58%, so anything above that is a genuinely close resemblance. Your photo is used for the search and then discarded; it is never stored or used for training. Want the details? Read how Ollie works, how the celebrity database was built, or the FAQ.