Methodology brief
Studio predicts the distribution of human visual attention over the first two seconds of free-view on a static image, and returns it as a pixel-level heatmap.
What it produces
For any uploaded image, Studio returns a probability map of where human attention is most likely to land during free-view: a per-pixel heatmap where warmer regions draw more attention and cooler regions are more likely to be passed over. The map is rendered back onto your original canvas so you can read it in context, at a resolution matched to your input.
What it learns from
The model is calibrated on real human attention data gathered across a wide range of real-world visual use cases — packaging, advertising creative, and dense visual layouts among them. The data used to calibrate the model is kept separate from the data used to validate it internally. Subject panels skew toward adult Western consumers; performance on panels outside this demographic has not been separately characterized. Enterprise customers can request a category-specific calibration study using their own recruited panel.
Derived metrics
Alongside the heatmap, the workspace reports time-to-first-fixation (TTFF) and an estimated fixation count. These are model-derived proxies inferred from the predicted attention map, not independent measurements, and have not been separately validated against in-lab fixation sequences. Treat them as directional ordering signals, not absolute measurements.
What the model is not
Studio is not a re-skin of classical center-surround saliency, and it is not a wrapper around a general-purpose foundation model. It is a purpose-built attention predictor calibrated to real human free-view behavior — not to semantic segmentation, depth estimation, or object detection.
Known limitations
Free-view, first two seconds only
The model predicts where the eye lands on a static image during the first two seconds of free-viewing. It does not model search tasks, reading order, brand recognition, semantic understanding, emotional resonance, or repeat-purchase behaviour. For impulse packaging and OOH this two-second prior is often the most important signal; for premium or considered categories it is one input among many.
Predictions, not measurements
Outputs are model predictions, not human eye-tracking measurements. Treat them the way a senior planner treats Nielsen BASES: a fast, useful checkpoint, not a replacement for primary research. Before a major launch, validate against an in-lab study. We can help arrange this on Team and Enterprise tiers.
Neutral canvas, not real shelves
Versus and Grid modes composite designs onto a neutral dark canvas at fixed scale. Results reflect attention on that canvas, not on a real planogram with adjacent SKUs, realistic shelf lighting, or packaging substrate variation. Real-world context templates (composite onto shelf or OOH environment photos) are planned for Q3 2026 on Enterprise.
Static images only
The current model accepts static PNG and JPG inputs. Motion creative (video, GIF, animated DOOH) is outside scope. For OOH, the model is most applicable to transit shelter, airport lightbox, and print formats where free-view dwell is the relevant exposure condition. Roadside large-format and digital OOH with motion involve different perceptual conditions not reflected in the current training distribution.
Validation is distribution-dependent
Our internal validation reflects performance on data held out from the model's calibration set. Performance on your specific product category, shelf conditions, or brand may differ. Enterprise customers receive a category-specific out-of-domain evaluation and a custom calibration study.
Terms of service
Last updated: May 2026
Acceptance
By accessing or using VisorLabs Studio (“Service”), you agree to these Terms of Service. If you are using the Service on behalf of an organization, you represent that you have authority to bind that organization.
Permitted use
You may use the Service to upload images and receive predicted attention analysis outputs for your own design and creative work. You may not use the Service to circumvent rate limits, reverse-engineer the model, or resell API access without a written agreement with VisorLabs.
Your content
You retain all intellectual property rights to images you upload. You grant VisorLabs a limited, non-exclusive license to process those images solely to provide the Service. We do not use customer uploads to train or improve models without explicit written consent. See the Privacy section for data handling details.
Model outputs
Predicted attention maps and derived metrics are outputs of a statistical model. They are not guaranteed to accurately reflect actual human eye-tracking data for your specific use case. VisorLabs makes no warranty that outputs are accurate, complete, or suitable for any particular decision. Use at your own judgement.
Accounts and billing
Monthly plans are billed in advance and cancel at the end of the current cycle. Annual plans are prorated and refunded on the unused portion. We reserve the right to suspend accounts that materially exceed fair-use thresholds without a contracted volume agreement.
Changes
We may update these Terms from time to time. Material changes will be communicated by email to account holders at least 14 days before they take effect. Continued use after that date constitutes acceptance of the updated Terms.
Contact
Questions about these Terms: [email protected]
Privacy policy
Last updated: May 2026
What we collect
- Account information: email address, name, payment method (via Stripe — we do not store card numbers).
- Usage data: analyses run, modes used, timestamps. Used for billing and product improvement.
- Uploaded images: processed in-memory for inference. See below for retention details.
Image retention
Uploaded images are processed in memory to produce the attention heatmap and metrics. They are not written to persistent storage on Studio and Team tiers. Enterprise customers can configure explicit data retention policies and dedicated storage under their contract.
Model training
We do not use customer-uploaded images to train or fine-tune the attention model without explicit written consent. The production model was trained on a curated dataset of real human attention data with appropriate licensing. Enterprise category-tuned models are trained on customer data only with a signed data processing agreement in place.
Sub-processors
We use the following categories of sub-processors: cloud compute (inference infrastructure), payment processing (Stripe), authentication (Supabase), and analytics (privacy-first, aggregated only). A full sub-processor list is available to Enterprise customers under NDA.
Your rights
You have the right to access, correct, or delete your account data. To exercise these rights, email [email protected]. We will respond within 30 days. For GDPR-covered data, we will respond within 30 days and honor deletion requests within the statutory period.
GDPR & CCPA
VisorLabs acts as a data processor with respect to uploaded images. Enterprise contracts include a Data Processing Agreement (DPA) covering GDPR Article 28 requirements. CCPA-covered users may request deletion of personal data at any time. EU data residency is available on Enterprise.
Contact
Privacy questions: [email protected]
Pre-release artwork
We understand that packaging and creative artwork is competitively sensitive. We are happy to sign your standard mutual NDA before any upload. Email [email protected].
Accessibility
VisorLabs is committed to making its website usable for as many people as possible, including people with disabilities. We aim to conform to the Web Content Accessibility Guidelines (WCAG) 2.1 Level AA as our target standard, and we work to improve accessibility on an ongoing basis.
Accessibility is an area of continual effort rather than a fixed endpoint, and some parts of the site may not yet fully meet every guideline. If you encounter a barrier, have difficulty accessing any part of this site, or need information provided in an alternative format, please contact us at [email protected] and we will work with you to provide the information or service you need.