Attractiveness test

What goes into the score?

The score is the arithmetic mean of four equally weighted photo measurements. All scores are bounded to 0–100. The weights are a design choice. This AI attractiveness rater uses landmarks and cheek pixels from your selected photo.

• Facial symmetry. The existing Midline detector averages five landmark group scores. Each group subtracts the mean reflected-pair error, divided by pair separation, from one and multiplies by 100.
• Proportion balance. We measure the visible forehead-to-brow, brow-to-nose and nose-to-chin spans along the detected face axis. The shortest divided by the longest, times 100, describes how similar these spans are. The forehead point is a mesh landmark.
• Skin lighting balance. We average encoded luma (0.2126R + 0.7152G + 0.0722B) in two small patches inside the cheeks, then divide the lower mean by the higher and multiply by 100. Patch radius is 3.5% of outer eye separation, rounded to the nearest pixel with a one-pixel minimum. Shadows, makeup and facial hair can affect this reading.
• Eye, nose and mouth alignment. We average the absolute side offsets of the outer-eye midpoint, nose tip and mouth midpoint from the fitted axis, divide by outer-eye separation, subtract from one and multiply by 100.

Choose a photo that is easy to measure

Face the camera with a relaxed expression and even light. Keep both cheeks, your forehead and chin visible. The detector can reject tiny faces, head turn and multiple faces. Larger images are resized to a maximum 1,600-pixel edge before measurement. Your photo is processed in this browser tab.

Questions about the test

This tool gives a composite photo feature score. The beauty symmetry test experiment shows how color edits changed geometry readings in a controlled photo example.