Glance

How Glance guesses where you look

Your eyes sample a scene. Glance learns from those patterns.

  1. Your eye sees a small sharp patch.

    Imagine a tiny spotlight. The fovea, near the center of your retina, gives roughly 2° of sharp vision. The rest helps you notice where to look next.

  2. Your eyes jump, then pause.

    The jumps are saccades, roughly 3–4 per second. Pauses called fixations often last 200–300 ms. People look differently when they have different goals.

  3. Researchers teach a model with gaze records.

    Recorded fixations show patterns across images and viewers. A model learns which image features tend to attract those fixations.

  4. Glance guesses a map for your image.

    An installed model runs locally and assigns relative attention across pixels. Darker teal in this illustration means a stronger guess. A map describes a pattern, not one person’s measured gaze.

Try one useful question.

“Does my main action compete with the image?” Draw an area around each, compare versions, then test the stronger hypothesis with people.

Interactive walkthrough: How Glance judges a design

Explore an in-app walkthrough driven by real components on a sample design: see the image load, laser scan, heatmap generation, look order trajectory, and visual clarity score.

Backed by peer-reviewed research

Glance models are built on published scientific literature and evaluated on standard benchmarks. Predictions estimate where human viewers tend to look during early visual viewing (~3 seconds); they are statistical estimates, not guarantees.

Read full citations, licenses, and bibtex in research credits →

Questions before you start

Is the gaze order a recording?

No. Without a scanpath model, Glance uses peak picking with inhibition of return and a center prior. This is an approximation, not a recorded sequence.

Can I use a saliency map to prove usability?

No. A saliency map estimates attention. Usability testing is needed to check whether people complete their tasks.