How your face report works

One photo, measured in your browser

After you preview your photo, start the analysis and sign in, MediaPipe Face Landmarker locates facial landmarks. Code uses those locations to calculate ratios and angles. A language model does not invent the measurements.

The checks look for a usable single face and help identify unsuitable pose, blur or obstruction. A successful check means the photo can be processed; it does not prove every landmark is perfect.

What the numerical score means

The scoring engine compares its measurements with reference ranges, turns them into individual scores, and combines them into eyes, midface, jawline and symmetry scores and an overall result. We use a pinned version of FreeHarmony’s scoring engine.

A higher result means a closer match to those chosen rules. It is not the percentage of people you are more attractive than, and the reference ranges are not a universal definition of beauty.

The report also shows eighteen explanatory proportions and angles. They are a separate measurement set, not eighteen extra scores added to the overall number. The experimental mirror-difference score is also separate.

What the AI adds

A vision-capable model receives a compressed copy of your photo and measured findings to describe your overall impression and individual features. This is a subjective visual interpretation, supported where possible by things you can inspect. It may still make mistakes.

Text is checked for contradictions and unsupported claims before it is displayed. The model’s opinions do not change the calculated score. The full report can add styling suggestions and a generated preview, which do not replace the original photo or its analysis.

What has and has not been established

The same landmark data and scoring configuration produce the same numerical result. Different photographs can produce different landmark data because of expression, camera distance, pose and lighting.

We have not established a validated attractiveness accuracy rate, a population percentile or a guarantee that different photos of the same person stay within one point. The public sample demonstrates the workflow; it is not a validation study.