Visual tests

A matchScreenshot step renders the page in the browser and compares it there with a baseline PNG, in the same run as the steps that led to the state. Rendering runs on the CPU and is deterministic, so baselines compare exactly by default. When the page's display list is unchanged since the baseline was recorded, the step passes without rasterizing at all.

The browsers bundled in @povio/lightning-panda are built with rendering enabled. A browser you build yourself needs the render feature (see Rendering: screenshots and PDF).

The three forms

import { css, label, matchScreenshot, screenshotMatchers } from '@povio/lightning-panda';

// 1. Steps inside a run
await page.run([
  label('New todo').fill('Buy milk'),
  matchScreenshot('home-filled'),                        // the viewport
  css('form').matchScreenshot('todo-form'),              // one element's box
  matchScreenshot('home-full', { fullPage: true }),      // the whole document
]);

// 2. One step on its own
const result = await page.matchScreenshot('dashboard', { maxDiffPixels: 20 });

// 3. A bun:test matcher
expect.extend(screenshotMatchers);
await expect(page).toMatchScreenshot('home');

page.matchScreenshot(name, opts) is one run with one step; it returns the step's result. toMatchScreenshot(name, opts) wraps it: it passes when the screenshot matches and fails with the mismatch message otherwise.

Names may use letters, digits, ., -, _, @ and / (for subdirectories), and must not contain ... Anything else throws when the step is built.

Options

Option	Default	Meaning
threshold	0	0 to 1: how far a pixel's channels may differ and still count as matching
maxDiffPixels	0	Differing pixels allowed
maxDiffPixelRatio		Differing pixels allowed, as a fraction (0 to 1) of the image
clip		{ x, y, width, height }: a rectangle of the document in CSS px
fullPage	false	The whole document, not only the viewport
dir	__screenshots__ beside the test file	Where the baselines live
timeout	step default	How long the render may take

clip, fullPage and an element (loc.matchScreenshot) exclude each other. The test file's directory is found from the call stack where the step is built.

Baselines on disk

For matchScreenshot('home') in e2e/todos.test.ts:

File	Contents
e2e/__screenshots__/home.png	The baseline image. Commit it
e2e/__screenshots__/home.json	{ "sceneHash", "width", "height" }: the display-list hash of the baseline's render. Commit it with the PNG
e2e/__screenshots__/home.actual.png	Written on a mismatch: what the page looks like now
e2e/__screenshots__/home.diff.png	Written on a mismatch: differing pixels in red

The actual and diff files are removed again when the step next passes. Add *.actual.png and *.diff.png to .gitignore.

The SDK sends a baseline to the browser once per connection and remembers its hash, so repeated comparisons against the same baseline do not resend the PNG.

The display-list hash

The browser builds the page's display list before it rasterizes. When its hash equals the sceneHash stored beside the baseline, the step passes at once with hashHit: true and no pixels are drawn or compared. A changed display list is rasterized and compared pixel by pixel, so a change that does not alter any pixel still passes.

First run and updating

A step with no baseline renders the page, writes the PNG and its JSON, and fails the test:

screenshot 'home': the baseline did not exist; wrote /path/to/e2e/__screenshots__/home.png

The run reports the first missing baseline it meets and stops writing there, so a run with several new screenshots writes only the first one. Write them all at once in update mode, which records every baseline the run reaches instead of comparing:

How	Where
LP_UPDATE_SNAPSHOTS=1 (also true, all, changed)	Any runner
bun test -u or bun test --update-snapshots	bun test
lightningpanda e2e -u	The suite runner

LP_UPDATE_SNAPSHOTS=1 bun test e2e/visual.test.ts

Update mode rewrites every baseline the run reaches, matching or not. Run it for the tests you mean to update, look at the changed PNGs in your diff tool, then commit.

When a screenshot does not match

A mismatch throws a ScreenshotMismatchError (a subclass of Error, exported from the package):

step 1 (matchScreenshot /path/to/__screenshots__/home.png): matchScreenshot '/path/to/__screenshots__/home.png': 474981 pixels differ (800x600); the actual image and the diff are in error.data
  actual: /path/to/__screenshots__/home.actual.png
  diff:   /path/to/__screenshots__/home.diff.png

import { ScreenshotMismatchError } from '@povio/lightning-panda';

try {
  await page.run([matchScreenshot('home')]);
} catch (e) {
  if (e instanceof ScreenshotMismatchError) console.log(e.result.diffPixels, e.result.files);
  throw e;
}

e.result is a ScreenshotResult:

Field	Meaning
name	The step's name
pass	Whether it matched
hashHit	Passed on the display-list hash, without rasterizing
recorded	No baseline was compared: the render was written as the baseline
diffPixels	Pixels that differ beyond threshold
sceneHash, width, height	Of the new render
ms	Browser-side timings of the render and compare
files	{ baseline, actual, diff } paths written

A successful run returns the same result per step in r.screenshots[name].

Look at the diff before updating a baseline. Use threshold or maxDiffPixels only for noise you understand, such as antialiasing on one edge.

What is captured

Captured:

• the live DOM after scripts ran, and every stylesheet, including ones added or edited by script
• the state of inputs, textareas and selects
• the window's scroll position, the viewport and the device scale factor
• images and fonts loaded through the page's network (mocks apply)

Not captured:

• canvas and video pixels
• iframe contents
• scroll offsets of elements other than the window
• :hover and :focus styles, the text caret
• running animations

Pixels come from Lightningpanda's own renderer (Blitz, rasterized on the CPU). They never match Chromium screenshots pixel for pixel: record baselines with Lightningpanda.

Baselines per operating system

The renderer uses the system's fonts for any text whose fonts the page does not load itself. macOS and Linux have different fonts, so their renders differ. Keep one set of baselines per OS, or load every font the page uses as a web font. A per-OS directory:

import { join } from 'node:path';

const dir = join(import.meta.dir, '__screenshots__', process.platform);
await page.run([matchScreenshot('home', { dir })]);

Record the Linux baselines on the same image your CI uses (for example ubuntu-24.04); see CI recipes.

Text that should wrap as Chrome wraps it between a hyphen and a digit needs args: ['--hyphen-digit-breaks', 'on'] on the server. It is off by default.

Playwright's toHaveScreenshot

The Playwright layer implements expect(page).toHaveScreenshot() with Playwright's own paths (<test file>-snapshots/<name>-<platform>.png). See Screenshots and snapshots.