Where to look on the result
The big number at the top is the fun score. It blends three signals, and the biggest part is mood match, which is about expressions. So two smiling strangers can get a high fun score without looking alike at all. For the look-alike question, skip the big number and look at two smaller things.
First, the face resemblance signal. It comes from a face recognition model and is the closest thing the tool has to "do these faces look like the same kind of face". Second, the Feature by feature bars. They compare five face ratios, one at a time, so you can see where the two faces line up and where they do not.
What the resemblance number means
The recognition model turns each face into a list of 128 numbers, called a face descriptor. Faces the model sees as similar end up with similar lists. The tool measures the distance between the two lists. A distance of 0.4 or closer counts as 100. A distance of 0.9 or more counts as 0. Everything in between is scaled in a straight line.
For a sense of scale, two photos of the same person usually land around 0.3 to 0.45. Two strangers usually land around 0.6 to 0.8. That gives rough bands you can use:
- 80 to 100: the model sees these faces as very close, near the range it gives two photos of one person.
- 60 to 79: more overlap than the model sees between most strangers. A fair claim to "you two look alike".
- 20 to 59: about what the model gives two strangers. This is where most pairs land.
- 0 to 19: the model sees two quite different faces. Totally normal, and no comment on anything else.
Reading the Feature by feature bars
The five bars compare face shape, eye spacing, nose length, mouth width and eyes to mouth. Each one is a ratio, so photo size and crop do not matter. A short gap means the two faces have nearly the same ratio for that feature.
The tool names the closest feature and the most different feature, and shows how far apart they are as a percentage. For example, "eye spacing 2% apart" means your eye spacing ratios are almost the same. "Mouth width 14% apart" means one of you has a wider mouth for your face, or one of you is smiling much bigger. Expression and head angle move these bars, so a fair photo matters.
How to run a fair look-alike test
A look-alike test compares two photos, so anything that differs between the photos shows up in the numbers. A few simple rules make it fairer:
- Both faces straight to the camera. A turned head changes the ratios more than anything else.
- Same light for both, ideally the same room and the same time of day.
- Same expression. Both neutral is best for a look-alike test, because big smiles change the mouth and eye ratios.
- No sunglasses, masks or face-reshaping filters. Clear glasses are fine.
Look-alike games to try
The test is a good excuse to dig through old photos. Here are a few rounds people enjoy:
- Sibling showdown: which two siblings does the model think look most alike? Use Group ranking with a family photo, and look at the Look-alikes highlight.
- Same age, different years: a parent's photo at 20 next to their child at 20.
- Couple then and now: your first photo together and a new one. Compare only the resemblance number, not the fun score.
- Match type hunt: if the result says Look-alikes, Different Moods, your faces match but your expressions do not. Retake with the same face and see if you become a Mirror Match.
What a resemblance test cannot tell you
The face model was built to tell people apart. It was not built to find relatives, and it cannot. A high number does not mean you are related, and a low one does not mean you are not. It is not an identity check either, so do not use it to prove who is in a photo.
It also says nothing about how well two people get on. If you are curious why some couples do look alike, the research page explains what the 1987 and 2020 studies found. For the full maths behind every signal, see how it works.