01 / Introduction

What HeatSight does.

HeatSight analyzes static designs and estimates where a viewer is likely to look first. Upload a poster, ad, screenshot, or mockup and get a heatmap, a numbered scan path, eight design scores, and a written AI report.

02 / Engine

How the pipeline works.

Your image is sent to a vision-language model through the AI Gateway. The model predicts likely attention points, scores the design across eight dimensions, and writes a short report. The heatmap and scan path are then rendered client-side from those predicted fixation points.

03 / Scores

Eight metrics, one composite.

Scores are model estimates, not lab eye-tracking measurements. They give you a structured way to compare versions and spot weak areas.

Hierarchy
How clearly primary → secondary → tertiary reads.
Contrast
Luminance and color separation between elements.
Typography
Type pairing, size ratios, and tracking.
Readability
Body text legibility at glance distance.
Balance
Weight distribution across the composition.
Spacing
Rhythm and negative space.
CTA visibility
Predicted first-second focus on the action.
Accessibility
Estimated color-vision and low-vision risk.
04 / Compare

A/B attention.

Upload two variants to see which composition routes the eye more predictably and gets the CTA read first. Compare mode uses the same model pipeline on both images.