Inside SZN

The method

Step 01

Guided capture

Face detection runs live in the browser with MediaPipe. Face count, framing, sharpness, exposure and lighting evenness are gated before we accept a photograph.

Step 02

Robust sampling

Skin, iris, brow, lip and hair samples are collected from multiple landmarks, filtered for specular highlights and shadow, then averaged with trimmed means.

Step 03

Deterministic colour math

Every sample is converted to CIELAB and LCh. Distances use CIEDE2000. Continuous temperature, value, chroma and contrast scores are derived from the measurements.

Step 04

Season classification

Twelve season prototypes are stored with versioning. Your profile is compared using a weighted squared distance, then softmax gives a probability distribution.

Step 05

Explanation, not decision

The language model never chooses your season. It may only put the numbers into elegant, human words.