face-consensus

Estimating the aggregate human rating of facial attractiveness

The model predicts the distribution of scores (1–5) that a documented population of 60 human raters assigned to a photograph in the SCUT-FBP5500 dataset. The number shown is the expectation of that distribution; the spread is the predicted disagreement among raters. It is a statistical aggregate of subjective preferences from that rater population — not an objective property of a person.

Drop a photograph or click to select.
A roughly centered face crop works best. Everything runs in your browser; nothing is uploaded.

Loading model…

Estimated mean rating of a specific rater population for this photograph, at annotation time. Photographs of the same person can score differently; the model rates the image, not the person.

These demo weights are untrained placeholders: the pipeline is real, the scores are noise. Export a trained checkpoint with scripts/export_onnx.py to replace them.