What makes two faces look alike?

We've all done a double-take at a stranger who could be someone's twin. Resemblance feels obvious in the moment — but what exactly is the eye (and the algorithm) picking up on?

Human faces vary along a huge number of subtle dimensions, yet a handful of structural features do most of the heavy lifting when we judge whether two people look alike.

The features that matter most

• Proportions, not just parts. The distances between features — eye-to-eye spacing, nose length relative to face height, the width of the jaw — matter more than any single feature on its own.
• Bone structure. Cheekbones, brow ridge and jawline define the underlying "scaffolding" of a face and stay consistent across weight changes and ageing.
• Eye and brow shape. We're extraordinarily sensitive to the region around the eyes — it's often the first thing that makes two faces feel related.
• Overall face outline. Round, oval, square or heart-shaped outlines set a strong first impression before we look at any detail.

Why the AI sometimes disagrees with you

People tend to anchor on hair, makeup and expression — exactly the things that change from day to day. A face-matching model like the one behind Dollganger is trained to ignore those and focus on stable structure. That's why it can pair two very different-looking photos as a strong match, or rate a "obvious" resemblance lower than you expected.

What throws it off

Extreme angles, heavy shadows, sunglasses and very low-resolution photos all remove structural information the model relies on. A clear, well-lit, front-facing portrait gives the most reliable results — the same conditions that help a human judge resemblance too.

Resemblance isn't relatedness

Finally, looking alike is not the same as being related. Similar facial geometry can appear across completely unconnected people — the world simply isn't big enough to give everyone a unique face. That's part of what makes finding a lookalike fun.