Frequently Asked Questions

VeysaLabs uses TranSalNet, a saliency prediction model that combines a convolutional neural network (CNN) encoder with transformer components. The transformer layers capture long-range spatial relationships in an image that CNN-only models miss, producing attention predictions that more closely reflect how humans actually scan visual content.

The model outputs a continuous saliency map: a spatial probability distribution showing where attention is likely to concentrate. This is rendered as a heatmap overlay on your creative.

The underlying research is published in Neurocomputing (2022). Read more on The Science page.

Saliency predictions achieve approximately 85 to 90% correlation with hardware-based eye tracking, validated against the MIT300 benchmark — the standard evaluation for saliency prediction models.

We are transparent about what synthetic eye tracking is: a fast, scalable, research-validated signal that correlates strongly with real fixation data at the population level. It is not a perfect replica of individual eye tracking sessions. For creative teams making decisions at speed and scale, it is a meaningful upgrade over instinct — and significantly faster and more accessible than lab-based testing.

TranSalNet was trained on the SALICON dataset (Saliency in Context), one of the largest publicly available human fixation datasets for saliency prediction. It contains approximately 10,000 images sourced from the Microsoft COCO image database, with fixation data collected from thousands of human participants via a validated psychophysical paradigm.

Each image was viewed by multiple people, so the heatmaps represent aggregated human attention — the statistical consensus of where people actually look, not a single viewer's response.

SALICON annotations are licensed under Creative Commons Attribution 4.0 (VIP Lab, University of Minnesota). TranSalNet is released under the MIT License. See the Open Source section below for full attribution.

Not yet. The current model predicts general population attention. This is a known limitation shared by all synthetic eye tracking tools available today.

Audience-segmented models are on our roadmap. We are exploring an academic research partnership to develop fixation datasets segmented by demographic and behavioural profiles. We will announce this when it is ready.

Yes. The Creative Effectiveness Report is grounded in four peer-reviewed studies:

Pieters & Wedel (2004) — attention capture and transfer in advertising, tested across 1,363 ads with over 3,600 consumers. Journal of Marketing Research.

Wedel & Pieters (2008) — a review of eye-tracking research in marketing. Review of Marketing Research.

Lou et al. (2022) — TranSalNet: towards perceptually relevant visual saliency prediction. Neurocomputing, 494, 455–467. DOI: 10.1016/j.neucom.2022.04.080

Xie et al. (2023) — audience-level visual attention and advertising effectiveness. IEEE Transactions.

Two ways in:

Image upload Upload a JPEG, PNG, or WebP file directly. Works for ads, social posts, packaging, OOH, landing page screenshots, pitch decks, or any visual asset.

URL scan Paste any live webpage URL and VeysaLabs automatically captures a full-page screenshot and runs the analysis. No screenshot needed on your side.

Video analysis is on our roadmap but not yet available.

Typically under 30 seconds from upload or URL entry to a complete heatmap and report. Inference runs on dedicated GPU infrastructure so results are near-instant regardless of image complexity.

A heatmap shows you where people look. A Creative Effectiveness Report tells you what to do about it.

The report includes visual saliency scoring, CTA and logo visibility analysis, gaze path prediction, clutter scoring, and specific research-backed recommendations to improve your creative before it goes live. The full report is available on Agency plans and above.

Agency adds Zone Tagging (define specific regions of interest), Scan Path animation (visualise the predicted sequence of eye movement), the full Creative Effectiveness Report, PDF export, 10 seats, 500 analyses per month, and priority support.

Starter includes heatmap overlay and analysis, 3 seats, 100 analyses per month, and priority email support.

Yes. Upload variants into the same project and compare attention metrics side by side. Available on all plans. Test A/B variants before media spend, not after.

No. Never. Your images are processed through our pre-trained saliency model and the results are returned to you. The model does not learn from your data. Your creative assets remain your property and are never used for any purpose beyond generating your requested analysis.

Your data is stored and processed across the following services, which are detailed in our Privacy Policy:

Replit — hosts the API, database (PostgreSQL), and image storage. Stores account data, project metadata, analysis results, uploaded images, and generated heatmaps. Based in the US.

Modal — runs GPU inference for the saliency model. Images are processed in memory only and are not persistently stored on inference servers. Based in the US (AWS/GCP).

All data is transmitted over HTTPS/TLS encryption. Passwords are hashed using bcrypt and never stored in plaintext. Access to production systems is restricted to authorised personnel only.

No. All data is isolated per customer at the database level. There is no shared access, no shared storage, and no cross-customer visibility.

Yes. Email privacy@veysalabs.io and we will send you our standard Data Processing Agreement within 24 hours. It covers GDPR and PDPA requirements.

Not yet. We are aligned with the ISO 27001 framework. Full certification is on our roadmap and will be pursued when an enterprise deal requires it. A full subprocessor list and security details are available in our Privacy Policy.

Yes. VeysaLabs complies with both the Singapore PDPA and EU/UK GDPR. We have a published Privacy Policy, a standard Data Processing Agreement available on request, and use Standard Contractual Clauses for international data transfers. Full details of our data handling and subprocessors are in the Privacy Policy.

The Beta plan gives you 1 seat with 25 analyses in total (no monthly reset) and includes heatmap overlay and analysis. It is designed to let you experience the full platform before committing. Beta access is available for a limited time only and is currently invite-only.

Yes. You can upgrade at any time and the change takes effect immediately. Downgrades apply at the next billing cycle. No lock-in contracts on any plan.

You will be notified as you approach your limit. Once reached, you can upgrade your plan or wait for the next billing cycle to reset. We do not charge overage fees without your explicit consent.

Enterprise plans start from $799/month and are customised based on team size, analysis volume, and integration requirements. This includes unlimited analyses, custom seats, everything in the Agency plan, and a dedicated account manager.

White-label, SSO, and audience segmentation features are on the roadmap. Contact us for a tailored quote.

Yes. The saliency prediction engine is built on TranSalNet, which is released under the MIT License. The training data annotations are licensed under Creative Commons Attribution 4.0. We are grateful to the researchers whose work makes this platform possible.

No, and we are transparent about this. The SALICON training images are sourced from the Microsoft COCO dataset, which in turn uses images from Flickr. Neither VeysaLabs nor the SALICON team owns copyright to those images.

Importantly, VeysaLabs does not distribute, reproduce, or expose any of these training images. We run inference only — your uploaded creative is analysed using pre-trained model weights. The training images are never transmitted, stored, or surfaced to users of the platform.