UI Verify

Change detection

An AI judge, not a pixel diff

A pixel diff flags every changed pixel, so you drown in noise or turn the threshold down and miss the real regression. UI Verify reads your pull request’s intent and tells an intended change from a real regression, with its reasoning, so you review decisions instead of diffs.

pages/Product
Regression
Baseline
Sunrise Mug 4.9

Stoneware · 12oz

$24

Add to cart
Current
Sunrise Mug 4.9

Stoneware · 12oz

$24

Add to cart

AI judge: regression

Confidence: high
Merge blocked

The Add to cart button lost its fill and text, so it reads as a faint outline instead of the store’s primary action. The pull request only asked to restyle the button, so this change was not the intent. The check stays red until a human accepts or the code is fixed.

Three verdicts, each with a reason

For every changed snapshot the judge makes one call and explains it, so a red build means a real change, not a threshold you have to babysit.

Intended

The pull request restyles the Add to cart button, and that new style is what changed. Accept it as the new baseline.

Regression

The restyle stripped the button's fill, so the Add to cart button is now a faint outline. This was not the intent.

Flake

Only a one-pixel anti-aliasing shift on a font edge, with no change to layout or content. Ignore it.

A verdict is only as good as the capture

Most visual-testing pain is flake: the same UI screenshots differently run to run, so the tool cries wolf and you stop trusting it. We pin the sources of that drift before the judge ever looks.

Animations, timers, fonts, and async data are all held steady, across Chromium, Firefox, and Safari at desktop and phone width. The sub-pixel noise nothing can pin, the judge labels a flake and keeps out of your review. So when a build comes back changed, it is a change worth reading.

Let your agent act on the verdict

Snapshot only what changed

We capture only what your change touches and carry the rest forward, so a typical PR re-renders a fraction of your UI and bills a fraction of it too.

Versioned baselines

Every accepted baseline is kept, so a verdict always compares against a real, reviewed reference and you can diff against any point in time.

Review decisions, not diffs

Start free in minutes, no credit card. Or book a demo and we’ll get you live in your CI.