[LOCAL SALIENCY] · ONNX RUNTIME · MIT CODE

See what gets the first glance.

Free attention heatmaps for design reviews. Predict early visual focus directly on your computer—no images ever leave your device.

  • Zero cloud upload
  • No account needed
  • Browser + desktop app
  • Windows + macOS · MIT code
glance // visual-attention-analysis offline // local-inference
Glance desktop app interface analyzing a web page design, showing predicted attention heatmaps, focus order, and area-of-interest scores
Real Glance interface: saliency heatmap and areas of interest computed locally on the device. Check download status →
Workflow

How it works in 3 steps

Form an attention hypothesis in seconds without setup or telemetry.

  1. Add your design

    Drop a screenshot or export (PNG, JPG, WebP) of your landing page, banner, UI element, or thumbnail.

  2. Compute attention locally

    Local vision models estimate where visual contrast, shapes, and color draw the first glance in early free viewing.

  3. Inspect focal hierarchy

    Review predicted attention hotspots, check competing elements, and verify your main call to action is seen first.

Analysis

Give your next review a starting point

Catch visual hierarchy flaws during design review before conducting user testing.

Fixation order

Focus order

Inspect likely attention around your headline, image, and main action before conscious reading begins.

Visual distraction

Find competing elements

Spot high-contrast secondary badges, background graphics, or clutter that fight against your primary action.

A/B verification

Compare one change

Draw areas of interest, then compare predicted attention shares between two versions before launch.

Local-first privacy

Your image stays on your computer

Glance is designed to run installed ONNX models locally. The first model download needs a connection. External integrations and MCP clients have their own privacy boundaries. No user images or telemetry are ever sent to an external server.

Review integration setup →
Peer-reviewed literature

Backed by published visual attention research

Glance models are built on peer-reviewed saliency publications and evaluated on independent benchmarks against split-half human eye-tracking ceilings. Predictions estimate where human eyes tend to land in early viewing (~3 seconds); they are statistical estimates, not guarantees.

UNISAL · Fast & Light

UNISAL

Fast and light. Good first check for any image. Unified lightweight model with domain subnets for image and video saliency.

Droste et al., ECCV 2020 →

TranSalNet · Global Context

TranSalNet

Sees the whole layout. Better on busy pages where context matters. Combines CNNs with transformer attention blocks for global structure.

Lou et al., Neurocomputing 2022 →

SUM · Cross-Domain

SUM

Tuned per content type: web pages, ads, e-commerce, photos. State-space modeling (Mamba) adapting to varied visual design domains.

Hosseini et al., WACV 2025 →

Open in-app guide: How Glance judges a design → · MIT/Tübingen Saliency Benchmark →
Scientific boundary

A prediction you can question

Predictive eye tracking estimates where people may look during early free viewing. It cannot tell you whether they understand your message, remember your brand, or buy your product. Use Glance to form an early attention hypothesis—not as a substitute for real usability testing.

Read the science and limits →
FAQ

Questions before you start

Clear facts about technology, privacy, and licensing.

Does Glance record my eyes?

No. Glance predicts likely attention from an image. It does not record gaze or use a camera.

Is Glance free?

Glance is a free, open-source desktop app in development. Its own code is MIT; model assets have separate terms.

Does Glance offer free AI eye tracking?

Glance is a free saliency map generator for predicting attention. Actual eye tracking requires recorded gaze; a prediction does not measure your eyes.

Can I use Glance as a free neuromarketing heatmap?

Use Glance to form an early attention hypothesis. It does not measure emotion, memory, purchase intent or conversion.

See what gets the first glance

Try UNISAL, SUM or TranSalNet in your browser. SUM and TranSalNet download once when you choose, then stay stored in this browser. The faster desktop app adds video and batch reviews.