How UI Analyzer scores a design

From screenshot to report

Uploads are limited to 10 MB and must be at least 200 × 200 pixels. The server rotates the image using its orientation metadata, resizes it to fit within 1024 × 1024 pixels without enlargement and compresses it to JPEG at quality 80. The processed image is stored on Cloudinary.

An AI vision pass observes the image using the selected design context. A separate text-only AI pass scores those observations and produces feedback. Both requests set temperature to zero, but that does not guarantee identical output. Scores are rounded and clamped to 0–100; the server combines them using the weights below. Recommendations, palette and available strengths are saved with the report.

Eight criteria

Color usage

Visual contrast, palette relationships and consistency of color use. These are AI observations, not a measured WCAG compliance audit.

Spacing

Apparent padding, gaps, grouping and breathing room in the screenshot.

Alignment

Visible edges, grids and relationships between elements.

Consistency

Repeated component styling and use of a coherent visual system.

Hierarchy

Relative emphasis and visual priority of content and actions.

Typography

Apparent type scale, readability, text density and visual text issues.

Corner rounding

Consistency and suitability of visible corner shapes.

Icon consistency

Icon family, stroke weight, size and color. Skipped when there are no icons or a usable icon score is unavailable.

Known limitations and evaluation

Advisory visual feedback, not a usability study or accessibility certification. Image quality, selected mode, omitted context and model changes can affect results. Historical percentiles compare stored scans, not an independent quality benchmark.

A static screenshot cannot test navigation, assistive technology, keyboard access, real touch-target dimensions or user task completion. Resizing may obscure small text and fine details. Contrast, spacing and pixel distances described by the AI are estimates; the active analysis path does not measure every pixel or read the source DOM.

Results may vary with screenshot cropping, compression, mode selection, provider/model changes and AI interpretation. The scoring weights and image/AI pipeline have been reviewed against the source implementation. No published accuracy benchmark or documented programme of manual design-review comparisons was found. Internal evaluation may use retained scans; this does not establish a measured accuracy rate.

Retention and product improvement

Screenshots are uploaded to our server, resized/compressed into JPEGs, stored on Cloudinary and sent to a third-party AI provider for analysis. Image URLs, mode, scores, feedback, recommendations, palette, timestamps and hashed IP addresses are stored in the database. Uploaded designs and analysis results are retained indefinitely to provide shareable results, maintain scan history and improve the accuracy and consistency of UI Analyzer. We may use submitted designs, scores and analysis results to evaluate patterns, test changes and improve the tool internally. Eligible scans may appear in Recent Scans/public areas and shareable result links may remain accessible. This internal evaluation is separate from third-party AI processing; it is not a claim that we or a provider train AI on submissions. There is no automatic expiry. For privacy or deletion requests, use the contact page and include your result link; removal of a database report alone does not remove its Cloudinary image.

Read the Privacy Policy or report an incorrect result or privacy concern.