How the look-alike test works
The test uses the same face comparison as the SightRadar API. Each face is turned into an embedding, a list of numbers that describes its features, and the two embeddings are compared. The result is a similarity score from 0 to 1. Two photos of one person usually score high. Two different people score lower, and how much lower depends on how alike they look.
It compares two faces you choose. It does not search for anyone, it does not name anyone, and both photos are discarded once the score is worked out.
How to read the result
The gauge marks the line the API uses to call two photos the same person, 0.42 by default. The tool reads your score against that line:
- 0.42 and above, easily mistaken for each other: face recognition would treat the two photos as one person.
- 0.28 to 0.42, a strong resemblance: close to the line. The two faces have a lot in common.
- 0.15 to 0.28, some resemblance: some shared features, well short of the line.
- Below 0.15, not much resemblance: the faces have little in common in these photos.
The bands are this tool's own reading, made for comparing different people. To check whether two photos show the same person, use face compare, which reads the same score as a match or no match.
What the score can and can't tell you
- It measures looks, not family. Relatives can score higher than strangers, and so can unrelated people who happen to look alike. The score can't tell those apart and is not a DNA test.
- It depends on the photos. Light, angle, expression, glasses and age all move the score. The same two people can score differently with another pair of photos.
- It is not a percentage of shared features. It is a measure of how close two faces are as the model sees them, on a 0 to 1 scale.
Getting a fair comparison
- Use one clear face per photo, large in the frame and facing the camera.
- Pick photos taken at a similar age, with similar light and a similar expression.
- Skip sunglasses, heavy filters and side-on shots.
- Compare only people who are happy to be compared.
Want to know which parent you take after? The mom or dad test compares your face with two others and shows which scores closer.
The same comparison in the API
This page runs POST /v1/compare, the SightRadar equivalent of AWS Rekognition CompareFaces. Products use it to match a selfie to an ID photo, to spot duplicate accounts and to confirm a returning user. On the API a comparison costs 186 credits (93 per image). See the compare endpoint reference and the thresholds guide.
Frequently asked questions
- How does the look-alike test work?
- Each face is turned into an embedding, a list of numbers that summarises its features, and the two embeddings are compared with cosine similarity. The result runs from 0 to 1. The closer to 1, the more the two faces have in common. It is the same measurement face recognition uses to decide whether two photos show one person.
- What do the results mean?
- The tool reads the score against the line the API uses to call two photos the same person. Above that line, the two could easily be mistaken for each other. Just under it is a strong resemblance, further down is some resemblance, and a low score means the faces have little in common. Most pairs of different people score well under the line, so a strong resemblance is unusual.
- Is the score a percentage?
- Not quite. Some tools show the same number as a percent, so 0.37 becomes 37%, but it is not the share of features two people have in common. It is a measure of how close two faces are as the model sees them, on a scale from 0 to 1.
- Which photos work best?
- One clear face per photo, large in the frame, well lit and facing the camera. Photos taken at a similar age and angle give the fairest comparison. Sunglasses, heavy filters and side-on shots lower the score. When a photo has several faces, the largest, most confident one is used.
- Are my photos stored?
- No. Both photos are processed in memory to work out the score and then discarded. No photo, face crop or faceprint is saved, and nothing is used to train models. The picture you can download of the result is made in your browser.
- Can I use anyone's photo?
- Use photos of yourself and of people who are happy to be compared. The tool does not identify anyone or search for a person: it only compares the two photos you give it.
- Is the look-alike test free?
- Yes. There is no sign-up and no card. A fair-use limit per visitor keeps the tool fast for everyone; if you reach it, the tool shows how long to wait.
More free tools
- Face compare Upload two photos and get a similarity score, a match verdict and a plain-English reading of it.
- Mom or dad test Upload a photo of yourself and one of each parent to see whose face scores closer to yours.
- Face blur Upload a photo and the faces in it are blurred for you. Keep the ones you want sharp, then download.
- Face pixelator Upload a photo and the faces in it are pixelated. Pick how coarse the blocks are, then download.
- Face crop Upload a photo and get a centred crop of every face: square, portrait or circle, one file each.
- Face detector Upload a photo to count the faces in it, see a numbered box on each one, and check which are clear enough to match.
- API cost calculator Price your monthly face workload on AWS Rekognition, Azure Face, Google Cloud Vision and SightRadar, operation by operation.
- Rekognition calculator Price an AWS Rekognition image workload in any region, tier by tier: face search, detection, storage and liveness.