How LooksCompare produces a face similarity estimate
This page explains the comparison process in plain language, what influences the result, and the limits users should keep in mind when reading a similarity percentage.
1. Face detection and image preparation
When a user uploads a photo, the system first needs to identify the visible face area. Clear, well-lit and reasonably frontal photos generally provide better input than images where the face is very small, blurred, strongly filtered or partly hidden.
2. Numerical face representation
The AI model converts visible facial information into a numerical representation. This representation is designed to summarize visual patterns learned by the model so that two uploaded faces can be compared mathematically.
3. Pair-by-pair comparison
In the family test, the child image is compared separately with the mom image and the dad image. Each pair receives its own visual similarity estimate. The site then presents the two results so users can see which uploaded parent photo appears closer to the child photo under those conditions.
4. What affects the score
A similarity score is not determined only by the people in the images. The photographs themselves matter. Camera angle, lighting, image sharpness, lens perspective, expression, age, facial hair, makeup, filters and cropping can all influence the visible facial information available to the model.
5. Why the result can change
If a user uploads a different photo of the same person, the percentage may move. That is normal. A photograph captures one moment and one view of a three-dimensional face. Different images provide different visual information.
6. Normal and strict modes
LooksCompare can provide different comparison modes. A stricter mode may interpret similarity more conservatively than a normal mode, which can lead to lower percentages. These modes should be treated as different scoring approaches, not as separate scientific truths.
7. How to interpret close results
If the two parent scores are close, the sensible interpretation is that the child looks broadly similar to both in the uploaded images. A tiny numerical difference should not be treated as a decisive finding, especially because another set of photos can change the gap.
8. What the score does not measure
The system does not measure DNA, biological parentage, personality, ancestry, medical traits or legal identity. A high score does not prove relationship, and a low score does not disprove it.
How to get a more stable comparison
- Use clear, front-facing or near-front-facing portraits.
- Choose similar lighting conditions for all photos.
- Avoid strong beauty filters and extreme edits.
- Keep the full face visible.
- Use reasonably similar camera angles.
- Try more than one good photo if you want to see how stable the result is.
Learn more
How AI face comparison works →