Local visual-regression checks after a small website edit

Choose Playwright Test for local screenshot checks on a small site. Its lead over WebdriverIO is narrow; TestivAI Witness adds element and style context, and Testplane offers graphical review.

PRACTICAL PICKSPreferredPickOur Preferred PickLocal checks for a small website$0

Playwright Test

One assertion compares a page or element with a stored baseline and fails the test on a difference or a missing baseline. Free, no account, maintained by Microsoft. Its lead over WebdriverIO is narrow (4.30 versus 4.03).

Make and compare baselines in one pinned environment. It shows which pixels changed, not why; review each image before accepting it.

Why Playwright Test is our Pick
Newcomer assessment

No Rising Star selected

We could not establish a newcomer’s complete eligibility for this local Windows and macOS workflow.

TestivAI’s adapter workflow is documented, but support for both operating systems remains unconfirmed.

Who is this article for? Developers maintaining a small website who want to catch unintended visual changes after CSS or layout edits, at desktop and mobile widths, locally and in CI, without an account.

About 18 min read

What Preferred Pick, Rising Star and Ready to Rock mean

The quick list

Start with Playwright Test for a new small-site setup. Consider WebdriverIO for an existing suite, TestivAI for element context, or Testplane for graphical review.

All four local packages are free and need no account, checked November 6, 2026. CI time, image storage and review time remain costs. We could not confirm the complete local cost of Vizzly.

The comparison at a glance

Compare how each tool fails a CI run, handles moving content and accepts changes, and how well each is maintained.

Playwright Test leads narrowly, followed by WebdriverIO, TestivAI Witness and Testplane. Order reflects our judgment for local small-site checks; it is not a measured detection or speed ranking.

Local screenshot checks for a small website at desktop and mobile widths
FeaturePlaywright TestWebdriverIO visual serviceTestivAI WitnessTestplane
Best fitPreferred Pick: most small sites, with one test fileExisting WebdriverIO suiteElement and style contextLocal visual assertions and GUI review
Our assessment (out of 5)4.30 · narrow lead over WebdriverIO4.033.953.90
Price and license$0; Apache-2.0; no account$0; MIT; no account$0; MIT; no account$0; MIT; no account
Fails the CI run on a visual changeAssertion fails the testUse a visual matcher or assert the percentage returned by check*Run testivai report --fail-on-diff; reporter success alone is insufficientVisual assertion fails the test on a difference
Missing baseline in CIFails by default; --update-snapshots=missing passes since 1.64Quick start auto-saves and passes; disable autoSaveBaseline in CIGate rejects missing captures by default; do not enable missing-baseline bypassIntroductory assertion fails when its reference image is missing
Main detection limitsPixel-only; environment-sensitive; user reports lazy-load flakinessMissing baselines can pass; comparison-engine changes alter resultsStyle digests omit some visual properties; keep noise approval offRunner configuration needed; no documented CSS-cause explanation in overview
Reader reviewsNo reviews yet. Write a reviewNo reviews yet. Write a reviewNo reviews yet. Write a reviewNo reviews yet. Write a review

Choose Playwright Test for a new small-site setup; its lead over WebdriverIO is narrow. All four local packages are free and keep baselines in your repository. We could not confirm the complete local cost of Vizzly. CI compute, storage and human review remain costs. The October 1, 2026 check covered one synthetic page on Linux before our research-only method; it did not decide these recommendations. No measured speed or reliability ranking is claimed.

All features and sources 14 criteria

Open the groups that matter to you. They compare the same selected tools. “Planned” means future work, not an available feature. Sources include the published conditions.

Capturing and comparing pages

Show 6 criteria
Capturing and comparing pages
FeaturePlaywright TestWebdriverIO visual serviceTestivAI WitnessTestplane
How you add a checkCall toHaveScreenshot() in a testInstall visual service; configure runner and visual matchersAdd Playwright adapter and capture call, then a separate report gateSetup wizard, runner configuration and assertView() in a test
Fails the CI run on a visual changeAssertion fails the testUse a visual matcher or assert the percentage returned by check*Run testivai report --fail-on-diff; reporter success alone is insufficientVisual assertion fails the test on a difference
Missing baseline in CIFails by default; --update-snapshots=missing passes since 1.64Quick start auto-saves and passes; disable autoSaveBaseline in CIGate rejects missing captures by default; do not enable missing-baseline bypassIntroductory assertion fails when its reference image is missing
Desktop and mobile widthsProjects or device presets (emulation)Browser dimensions and mobile emulationPlaywright projects or device emulationConfigure browser dimensions for desktop and mobile layouts
Stable capturesWaits for two matches; animations stopped and caret hidden by defaultDisable CSS animation; control page readiness in test codeMaker: freeze animations, hide caret and await fonts by defaultMaker documents automatic waits and retries; pin the rendering environment
Changing contentMasks and a capture stylesheetHide, remove or ignore elements and regionsSelector ignores and masksExact masking and dynamic-content options not established in the reviewed overview
Sources for capturing and comparing pages

Playwright visual comparisons and snapshot modes (checked October 19, 2026); WebdriverIO repository, TestivAI repository, Testplane README (checked November 6, 2026). A missing description is unresolved, not evidence of absence.

Review, baselines and upkeep

Show 8 criteria
Review, baselines and upkeep
FeaturePlaywright TestWebdriverIO visual serviceTestivAI WitnessTestplane
What the report showsExpected, actual and difference imagesActual and difference images; optional visual HTML reporterHTML and JSON; DOM, style and selector contextLocal GUI and HTML reporter with several diff modes
Accepting changesReview then --update-snapshotsReview then --update-visual-baselineReview then testivai approve; optional PR approvalReview and accept reference images in the GUI
Baseline filesPNG or lossless WebP beside testsLocal baseline directoryPNG, DOM and element files in your repositoryLocal reference images
Release evidence1.64.1, October 14, 2026 (checked October 19)Visual service 11.0.0, October 10, 2026Core 2.7.0; Playwright adapter 2.1.1, November 4, 20269.3.6, November 2, 2026
Maintenance and compatibilityMicrosoft team; roughly monthly releases at October checkActive v11 line; WebdriverIO 10 requiredBudi Sugianto; local portable files, no support commitmentActive public contributors; exact dependency support lifecycle unknown
Security evidenceBrowser installer advisory fixed in 1.55.1 (CVE-2025-59288)No published repository advisories; dependency audit unavailableNo published repository advisories; optional MCP dependency fix in NovemberCurrent dependency advisory status not established
Main detection limitsPixel-only; environment-sensitive; user reports lazy-load flakinessMissing baselines can pass; comparison-engine changes alter resultsStyle digests omit some visual properties; keep noise approval offRunner configuration needed; no documented CSS-cause explanation in overview
AI features (not scored)Optional Claude Code test agentsNot needed for local comparisonsOptional Claude triage, MCP and Claude Code reportsNo AI feature scored
Sources for review, baselines and upkeep

Playwright release notes and advisory (checked October 19, 2026); WebdriverIO releases, TestivAI releases, TestivAI maintenance, Testplane release and project history (checked November 6, 2026). AI features carry no weight.

What matters when choosing

Playwright Test’s screenshot assertion is our Practical Pick for a developer who wants local checks that catch unintended visual changes on a small website after a CSS or layout edit. It is free under the Apache 2.0 license and needs no account. One line, await expect(page).toHaveScreenshot(), captures a page or an element, compares it with a baseline image stored beside your tests and fails the test when the two differ. Its documentation says it waits until two consecutive screenshots match before comparing, and it stops animations and hides the text cursor by default. A missing baseline also fails the test by default, so a CI run cannot pass without one. At our October 19 check, Microsoft maintained it with roughly monthly releases; version 1.64.1 had come out on October 14, 2026. You do not need an existing test suite: one test file that visits your pages at a desktop and a mobile size is enough to start. Visual comparisons, assertion options, snapshot update modes, license

The most important compromise is that a baseline belongs to the environment that made it. Playwright warns that the operating system, browser version, settings, hardware, power source and headless mode can all change a screenshot, so create and compare baselines in the same place, for example its official Docker image in CI. It compares pixels only: it shows expected, actual and difference images, but it does not say which CSS rule caused a change. Since version 1.64, running with --update-snapshots=missing lets tests that only create new baselines pass, so keep CI on the default mode if a missing baseline should fail the run. Visual comparisons, Docker, release notes

TestivAI Witness adds element and style context to a pixel diff. Its maker documents local screenshots paired with DOM and computed-style information, so the report can point to changed elements. Use its separate testivai report --fail-on-diff command to fail CI; a successful Playwright reporter run alone does not establish that the images match. It is free, needs no account and keeps baselines in your repository. Its main limits are the extra gate to configure and incomplete style coverage: an unchanged style digest does not rule out every visual defect. Repository, maintenance (checked November 6, 2026).

Playwright leads WebdriverIO narrowly, 4.30 to 4.03 out of 5. WebdriverIO is a sensible alternative if you already use it, but starting a new suite adds configuration and browser upkeep. TestivAI scores 3.95; Testplane scores 3.90. These are editorial judgments for this reader, not measured detection rates.

All four tools are free and keep baselines in your repository. Your costs are CI time, storage for images and the time it takes to review each change before you accept a new baseline. We could not confirm the complete local cost of Vizzly. These checks show that rendering changed; they cannot tell you whether the new design is right. Hosted services such as Argos, Chromatic and Percy need an account and upload screenshots, so they are outside this comparison.

Disclosure: Practical Picks articles are researched and written with AI, currently Anthropic’s Claude. Playwright documents optional test agents that can run with Claude Code; TestivAI offers optional Claude triage and a Claude Code report integration. We have no relationship with these makers or Anthropic over this article, and these integrations carried no weight in our judgment. Playwright test agents, TestivAI repository (TestivAI checked November 6, 2026).

Playwright Test

Preferred Pick · local checks for a small website

$0Apache-2.0 · no account

Choose its screenshot assertion for local visual checks on a small website, with or without an existing Playwright suite. Make and compare baselines in one pinned environment, and review each changed image before accepting it.

Reasons to choose it

  • Waits for a stable capture; animations off and caret hidden by default
  • Fails the test on a mismatch, and on a missing baseline by default
  • Maintained by Microsoft with roughly monthly releases

What to watch

  • Baselines depend on OS, browser version and hardware
  • Pixel differences do not identify the CSS cause
  • Since 1.64, a run with --update-snapshots=missing passes when it only adds baselines

The useful unit here is the screenshot assertion, not a claim that Playwright wins every browser-testing job. A test opens a page at a fixed size, and toHaveScreenshot() compares it with a named baseline. The first run writes the baseline and fails; later runs fail when the image differs beyond your tolerance and keep expected, actual and difference images for the HTML report. Baseline file names include the browser and platform, so a Linux baseline does not overwrite a macOS one, and they are stored as PNG or, since version 1.62, lossless WebP. You can cover desktop and mobile widths as separate projects or device presets. Mobile here means browser emulation, not a physical phone. Visual comparisons, emulation, release notes

To accept intended changes, run the tests with --update-snapshots. Without a value it updates only the screenshots that no longer match; all, missing and none are the other modes. Review each changed image before you commit it, because an unintended change captured in the same run is accepted with the intended one. Command line, snapshot update modes

Read the full Playwright Test analysis and sources

Controls for content that moves, and their risk

Playwright compares images with the pixelmatch library. threshold sets how different a pixel’s color must be to count, from 0 (strict) to 1 (lax), with a default of 0.2; maxDiffPixels or maxDiffPixelRatio set how many differing pixels you accept. mask covers chosen elements with a solid box, and stylePath applies a stylesheet during capture, for example to hide a clock or an embedded video. These controls help with content that changes on its own. Masking too much, or setting loose tolerances, can hide the very defect you wanted to catch. Assertion options, visual comparisons

Know the environment, the browsers and the record

Playwright’s Firefox and WebKit are patched builds; its documentation says it does not work with the branded Firefox or Safari your visitors install, so a WebKit baseline is not a Safari screenshot. Each Playwright version needs specific browser builds, and the visual-comparison guide warns that browser version changes screenshots, so expect to review some differences after an upgrade. OSV lists one advisory for the playwright package: before version 1.55.1, released in October 2025, macOS scripts that reinstall branded Chrome and Edge downloaded installers without checking the server’s certificate (CVE-2025-59288). Browsers, advisory

In an earlier hands-on check on October 1, 2026, before we adopted our research-only method, the assertion in Playwright Test 1.62.1 passed unchanged desktop and mobile repeats of one synthetic local page and failed each of four deliberate CSS changes, including a 7 px padding change and a hidden badge, in Chrome for Testing 151 on Linux. That single page is consistent with the documented behavior. It does not measure reliability on real sites, and it did not decide this Pick. Earlier check record, October 1, 2026

The strongest reason not to choose it is that you write a little JavaScript or TypeScript test code. If you need a runner driven only by a configuration file, BackstopJS offers that approach, with a substantial maintenance warning below.

Used Playwright Test? Share your experience to help other readers choose.

WebdriverIO visual service

Local alternative · existing WebdriverIO suites

$0MIT · no account

Choose it if you already use WebdriverIO. Assert the returned difference or use a visual matcher, and disable automatic missing-baseline creation in CI.

Reasons to choose it

  • Viewport, full-page and element capture
  • Local baseline files and optional HTML reporter
  • Maintained service with documented version support

What to watch

  • More configuration for a new small-site suite
  • autoSaveBaseline in the quick start lets missing baselines pass
  • New comparison engines can require reviewed baseline updates

WebdriverIO’s MIT-licensed visual service captures viewports, full pages or elements and compares them with local baselines. Use a visual matcher such as toMatchFullPageSnapshot() to assert the result: a bare check* method returns a mismatch percentage. It supports desktop and mobile-sized browser captures, hiding elements and disabling CSS animation. Its optional visual reporter produces HTML reports. Repository, visual-testing documentation (checked November 6, 2026).

The quick-start configuration sets autoSaveBaseline: true, which saves a missing baseline and lets the test pass. Turn that off in CI when absent baselines should fail, and keep --update-visual-baseline for deliberate approval. Version 11 supports WebdriverIO 10; its new comparison metric can change mismatch percentages, so an upgrade needs human review of baselines. The repository’s support table and older installation paragraph disagree about the default major version: pin the version you intend to use. Recent reports also describe frame captures and Android Chrome problems with WebDriver BiDi; these do not establish a failure in ordinary desktop page capture. Repository and compatibility notes, release notes, issue list (checked November 6, 2026).

Read the full WebdriverIO visual service analysis and sources

Choose it when WebdriverIO is already your browser runner. For a small site starting from scratch, Playwright’s built-in assertion takes fewer parts to assemble.

Used WebdriverIO visual service? Share your experience to help other readers choose.

TestivAI Witness

Local alternative · element and style explanations

$0MIT · local core, no account

Choose TestivAI for element and computed-style context alongside a screenshot diff. Configure its separate CI gate and review changes before approval.

Reasons to choose it

  • DOM and style context through the Playwright adapter
  • Local HTML and JSON reports; baselines in your repository
  • Named maintainer and portable MIT-licensed code

What to watch

  • A successful reporter run alone is not the visual CI gate
  • Style digests do not cover every visual change
  • One maintainer; no support commitment

TestivAI pairs each screenshot with DOM information and, through its Playwright adapter, computed styles and an element map. Its report distinguishes structural changes, style changes and possible rendering noise, with selectors to help explain where a difference sits. The maker documents frozen animations, a hidden caret, waiting for fonts and selector masks. These are documented capabilities, not a measured reduction in false alarms. Repository (checked November 6, 2026).

The local core is MIT licensed, requires no account and uploads nothing to its maker. Budi Sugianto is the named maintainer. The maintenance page dates the core’s first release to March 20, 2026; the Playwright adapter was already public on January 9, 2026. Both fall within the past two years. Core 2.7.0 and Playwright adapter 2.1.1 were released on November 4. Maintenance, license, releases, privacy (checked November 6, 2026; January adapter date from our October 19 research).

Read the full TestivAI Witness analysis and sources

Keep testivai report --fail-on-diff as a separate CI gate and review every changed image before approving it. Our October 19 documentation research identified fixes for a page-size false pass and a missing-baseline bypass; those fixes are included in later releases. Style comparison also has limits: the repository warns that matching digests do not cover every size, overflow, transform or image change. Leave automatic noise approval off unless you understand its consequences. The standalone crawler does not collect the element map, so it does not provide the Playwright adapter’s style attribution. Repository, release history (checked November 6, 2026).

Choose TestivAI when identifying changed elements is worth extra setup. One developer maintains it, with no support commitment; its portable local files reduce shutdown exposure but cannot guarantee future compatibility. The security page lists no published advisory, which does not establish that every dependency is safe. Optional GitHub reporting and hosted AI clients can send results outside your machine; neither is required for the local comparison. Maintenance, security, privacy (checked November 6, 2026).

Used TestivAI Witness? Share your experience to help other readers choose.

Testplane

Local alternative · visual assertions and GUI review

$0MIT · no account

Choose Testplane for local visual assertions with a graphical review workflow. Keep reference and comparison environments consistent.

Reasons to choose it

  • Page and element visual assertions
  • Missing reference fails in the introductory example
  • Local GUI approval and multiple diff modes

What to watch

  • Adds runner and browser configuration
  • Reviewed overview does not establish CSS-cause explanations
  • Current dependency security status not established

Testplane, formerly Hermione, is a browser-test runner with screenshot assertions. Its maker documents local execution, page and element captures, automatic waits and retries, and a graphical interface for reviewing differences and accepting reference images. The introductory assertView() example fails when its reference image is missing and passes after you accept that image. It is MIT licensed and needs no hosted review service. Repository and setup example (checked November 6, 2026).

Choose it if you want a runner with visual review built into its workflow. Its setup wizard generates configuration and a test file; you still write code to visit the pages and choose what to capture. Use fixed browser dimensions for desktop and mobile layouts, and keep the browser, fonts and operating system consistent between reference and comparison runs. Starting a small suite adds more runner configuration than Playwright’s built-in screenshot assertion. The documented report offers several diff modes, but it does not establish TestivAI-style explanations of the CSS or element responsible for a change. Repository (checked November 6, 2026).

Read the full Testplane analysis and sources

The repository lists version 9.3.6 from November 2, 2026 and active contributors. It explicitly describes the change from Hermione as a rename, with the existing version sequence retained. Published source and local reference images provide an exit path; they do not guarantee future compatibility. We have not established the current security status of every dependency or measured its false-pass rate. Release and project history (checked November 6, 2026).

Used Testplane? Share your experience to help other readers choose.

Which product are we missing?

Know a product that belongs in this article? Tell us what it does differently and why it deserves a closer look. Newcomers welcome.

Readers and product teams can suggest candidates. Suggestions inform our research; they do not guarantee inclusion or a recommendation.

Suggest a product

Opens an email draft with this article attached as a link. Nothing is sent until you send it.

No email app? Send the product name, website, and this article’s title to admin@practicalpicks.guide.

What people are saying

Practical Picks has no reader reviews for these tools yet. In January 2026 the author of a small static site built with Astro described adding Playwright screenshot tests with one test file that visits each page and calls toHaveScreenshot(), keeping the images in the site’s repository and running the tests by hand after risky CSS changes. The author put the setup at about two hours and reported that lazy-loaded images made full-page captures flaky until the test scrolled through each page first. In the Hacker News discussion of that post, another developer running similar checks in CI said Playwright “regularly fails by timeout”. These are single accounts, not measurements. For BackstopJS, issue #1607 reports a missing button passing with requireSameDimensions: false; keeping the default size check on caught it. The maintained plugins also have public issue reports, but we found no comparable measurement of their false-pass rates. BackstopJS issue #1607 (checked October 19, 2026). Visual regression tests for my website, discussion, BackstopJS issue #1607

How we researched

This is public-source research. For this comparison we did not install or run the tools, create accounts or contact a maker. No maker supplied access or payment, and our links carry no affiliate tracking. We first compared these tools and one newer tool on October 1, 2026, including a small hands-on check on a synthetic page. On October 19, 2026 we rechecked the documentation, releases, maintenance records, issue reports and security advisories, and made the decision again from public evidence.

We wrote this for a developer maintaining a small website who wants to catch unintended visual changes after CSS or layout edits at desktop and mobile widths, locally and in CI, with baselines kept in the repository and no account. Before scoring, we fixed five criteria and their weights: detecting a change and failing the CI run, including on a missing baseline (30%); setup and upkeep (20%); the review and baseline-approval workflow (15%); maintenance and continuity (25%); and licence, account requirements, baseline ownership and exit path (10%). Playwright Test scored 4.30, WebdriverIO 4.03, TestivAI Witness 3.95 and Testplane 3.90 out of 5. Playwright’s lead over WebdriverIO is narrow.

Simon Smith’s maintained Cypress image-snapshot rewrite scored 3.85. It integrates capture and comparison in an existing Cypress suite; enable requireSnapshots=true so a missing baseline fails in CI. SnapDrift and Frontend Promotion Guard each scored 3.65: both document local screenshot gates, but SnapDrift needs strict failure configuration and FPG needs a system browser and configured CSS/style checks. Cypress rewrite, SnapDrift, Frontend Promotion Guard (checked November 6, 2026).

BackstopJS scored 3.08. Its configuration-file workflow and browser approval report remain useful, but its October 19 record showed no release since September 2024, a maintainer request and an acknowledged false-pass report when the size check was disabled. We would not start with it now. Cypress Visual Regression scored 3.73: it has local capture, pixel comparisons and separate baseline and regression modes, but its setup and review workflow offer less help for this reader than the leaders. BackstopJS (checked October 19, 2026), Cypress Visual Regression (checked November 6, 2026).

We retained the October 19 exclusions for Lost Pixel (sunset and archived), reg-suit and jest-image-snapshot (no integrated page capture), Loki and Vitest’s screenshot feature (component-focused), and Visual Regression Tracker (requires a review server). The original Cypress image-snapshot package is distinct from Simon Smith’s maintained rewrite. Hosted Argos, Chromatic, Percy, Applitools and Happo do not meet the no-account, local-baseline requirement.

On November 6, 2026 we examined candidate repositories, documentation and launch records for newer options. No Rising Star selected: we could not establish every eligibility requirement for a newcomer serving this reader on both Windows and macOS. TestivAI’s documentation supports its local adapter workflow, but does not establish both operating systems. Other candidates lacked a verified first-release date, complete platform evidence or a documented local CI gate. An unknown launch date does not remove a useful alternative from this comparison. We could not establish complete scores for ShotSweep, snap_diff-capybara or Vizzly: their missing-baseline behavior, released documentation or unattended local gate remained unclear. We did not treat missing independent reviews as a reason to reject a product.

ToolPublic sourcesWhat remains uncertain
Playwright TestVisual-comparison guide, assertion and configuration reference, command line, release notes, browsers and Docker pages, license, OSV advisory, npm release record, user report and discussion.How it reports a page whose height changed; false-pass and flake rates on real sites.
WebdriverIO visual serviceLocal capture documentation, repository, license, releases, maintenance, security page and issue list; checked November 6, 2026.The effect of frame and browser-protocol issues on an individual suite; baseline changes after an upgrade.
TestivAI WitnessRepository, MIT license, releases, named-maintainer page, terms, privacy, security and issue pages; checked November 6, 2026.Windows/macOS support, general attribution accuracy, dependency risk and false-pass rates; style digests do not cover every visual change.
TestplaneRepository setup example, visual assertions, local review workflow, MIT license and release history; checked November 6, 2026.Dependency security, exact support lifecycle and false-pass rates.
Newcomer eligibilityPublic release and workflow documentation examined November 6, 2026.No Rising Star selected; no newcomer had every requirement established.

The October 1 check covered one synthetic page with two plan cards at 980 × 700 and 390 × 844 pixels in one Linux browser build; mobile was emulation. It is kept as dated history and did not decide the Pick. Screenshots, reports and traces show whatever the page displays, so keep test pages free of secrets and control who can open CI artifacts. Nothing here establishes speed, false-positive or false-pass rates, or behavior on dynamic production pages. Next substantive review is due January 17, 2027, or sooner if a release or maintainer status changes. Found a changed release or a fix? Contact the editors.

Sources and evidence

Checked October 19, 2026:

November 6, 2026 additions: WebdriverIO repository, documentation, releases, security, issues; TestivAI repository, license, maintenance, releases, terms, privacy, security; Simon Smith plugin, releases, security, issues; Cypress license, security; Cypress Visual Regression; refreshed Playwright license and BackstopJS license.

Additional November 6 sources: Testplane local assertions, review and release history, SnapDrift local CLI and failure modes, Frontend Promotion Guard local verify workflow.