Lens catches the visual bugs functional tests miss. It diffs every checkpoint against its approved baseline, filters out rendering noise, and explains each real change, all inside your private SBOX, self-hosted or managed as a single-tenant instance dedicated to you.
Every checkpoint in a build lands in one queue, grouped by test and browser. Compare it to the baseline, decide, and move on.
Naive screenshot tools flag every anti-aliased edge, and reviewers learn to click Approve all. Lens settles the obvious cases on its own, so people only see changes that need a judgement.
A red blob tells you something moved. Lens AI tells you what moved, why it probably moved, and whether it broke accessibility.
Each build carries its branch, commit and browsers. A feature branch compares against its own approved screens and falls back to main when it has none.
Turn on e34:lens and drop a checkpoint wherever a screen matters. The hub takes the screenshot. Then let CI ask whether the build is clean.
// 1. Turn Lens on for the session options.setCapability("e34:token", token); options.setCapability("e34:lens", true);options.setCapability("e34:branch", "feature/send-money-rename"); // 2. Checkpoint any screen driver.executeScript("sbox_executor", Map.of( "action", "visualSnapshot", "arguments", Map.of( "name", "Account overview", "viewport", "1280x800")));
// Session connected to SBOX with lens=true await page.goto("https://app.internal/accounts"); await page.evaluate("sbox_executor", { action: "visualSnapshot", arguments: { name: "Account overview", viewport: "1280x800", fullPage: true, }, });
visual-gate:
stage: verify
script:
- npm run test:e2e
# saves the buildId returned by visualSnapshot
- curl --fail -H "Authorization: Bearer $SBOX_TOKEN" "$SBOX_URL/e34/api/visual/builds/$(cat lens-build-id)/status"
Hosted visual testing sends every screen of your app to someone else's cloud. Lens runs on the same private grid as your tests.
Run Lens in your own network, or on dedicated servers Element34 manages for you. Never shared with other customers.
Images go to your S3-compatible storage, including MinIO and Dell ECS.
Point AI at an internal gateway and screenshots never leave the network.
Autoscaling workers in the SBOX Helm chart, non-root and PodSecurity-ready.
Retries and a dead-letter list. A worker restart doesn't lose a checkpoint.
Trust internal CAs, so TLS-intercepting proxies don't break comparisons.
Every checkpoint records who reviewed it, when, and how.
No. Set e34:lens on the session and call the sbox_executor action visualSnapshot where a screen matters. With e34:lensOn: "navigate", Selenium sessions checkpoint every page load with no extra calls.
It excludes anti-aliased pixels, skips identical screens by perceptual hash, applies a noise-floor threshold, and supports ignore regions, per-region thresholds and four match levels. With Lens AI, a model can approve pure rendering noise, but only on an unambiguous yes.
Only if you configure it that way. Lens AI sends the baseline and changed screenshot to the provider endpoint you set, which can be an OpenAI-compatible gateway inside your network. Without Lens AI, comparison is fully local.
Selenium 3, Selenium 4 and Playwright sessions on SBOX, across the browsers your grid runs, including Chrome, Firefox and Edge. Every checkpoint records its browser and version.
Lens is an SBOX add-on, and Lens AI is a separate add-on on top of it. Both follow Element34's annual licence model, so cost doesn't depend on how many screenshots you compare.
Send one regression suite. We'll switch on Lens on your private grid, run two of your builds, and show you every diff it catches and every one it filters as noise.