Browser support
Chromium, Firefox and WebKit — the versions MotionPR is tested with, what works in each engine, how engines differ, and why every engine has its own baseline.
On this page
MotionPR captures with Playwright's builds of the three browser engines. Each engine is a first-class target: it has its own baseline, captures, comparison and report.
| Engine | Build tested | Install | Config / CLI |
|---|---|---|---|
| Chromium | Chrome for Testing 158.0.8127.18 (Blink) | npx motionpr install chromium | --browser chromium |
| Firefox | Firefox 159.0 (Gecko), Playwright build 1572 | npx motionpr install firefox | --browser firefox |
| WebKit | WebKit 27.2, Playwright build 2388 | npx motionpr install webkit | --browser webkit |
These are the builds that match MotionPR's bundled Playwright (1.65). npx motionpr install
downloads exactly these versions; other versions of an engine are not covered by MotionPR's
tests. Captures record the version the browser reports at runtime, never one taken from the
configuration.
Chromium is the default (browser.name). Pick another engine in the config, per run with
--browser, or for every command with MOTIONPR_BROWSER:
npx motionpr install firefox webkit
npx motionpr run --browser firefox
npx motionpr run --browser all # Chromium, Firefox, then WebKitWhat works in each engine
Everything below is covered by automated tests that run in each engine against a real browser (CI runs the browser suites once per engine).
| Area | Chromium | Firefox | WebKit |
|---|---|---|---|
| CSS transitions (multiple properties, interrupted, reversed) | Verified | Verified | Verified |
| CSS animations (iterations, direction, fill, keyframes) | Verified | Verified | Verified |
| Web Animations (reversed, cancelled, finished, delayed) | Verified | Verified | Verified |
| Deterministic replay (identical frames on repeated runs) | Verified | Verified | Verified |
| GSAP adapter (tweens, timelines, labels, staggers, reverse) | Verified | Verified | Verified |
| GSAP ScrollTrigger (metadata and live progress) | Verified | Verified | Verified |
| Anime.js v4 adapter (timelines, staggers, WAAPI, eases) | Verified | Verified | Verified |
| Lenis adapter (scroll motion, velocity, reduced motion) | Verified | Verified | Verified |
| GSAP with Lenis in one page | Verified | Verified | Verified |
| The demo app's 11 intended changes found, nothing else | Verified | Verified | Verified |
| CLI: capture, compare, run, baseline, report, doctor | Verified | Verified | Verified |
| CSS scroll-driven animations | Metadata | Metadata | Metadata |
The test counts per engine are in the support matrix.
Every engine has its own baseline
Engines lay out text, round geometry, rasterize pixels and schedule frames differently. A Firefox capture compared with a Chromium baseline would report those differences as if your code had changed. So MotionPR keeps everything per engine:
.motionpr/
├── baseline/
│ ├── chromium/ capture.json + assets/ (commit these)
│ ├── firefox/
│ └── webkit/
├── captures/<browser>/ recent captures and LATEST
├── comparison/<browser>/
└── report/
├── index.html one row per engine
└── <browser>/ that engine's report- A run compares each engine with its own baseline. A Firefox run never reads the Chromium baseline; when Firefox has none yet, the run says so and writes a capture-only report.
baseline accept --browser firefoxwritesbaseline/firefox/only. A capture can only become the baseline of the engine it was taken with.- Comparing two engines is possible, but only when you ask for it: cross-browser comparisons.
Baselines from MotionPR 0.1 (baseline/capture.json) are read as the Chromium baseline;
npx motionpr baseline migrate moves them into baseline/chromium/.
How the engines differ
These differences were measured while adding Firefox and WebKit. MotionPR handles each one; they are listed so that you can read captures from each engine correctly.
| Difference | Chromium | Firefox | WebKit | How MotionPR handles it |
|---|---|---|---|---|
performance.now() resolution | 0.1 ms | 1 ms | 1 ms | Recorded as timerResolutionMs; a COARSE_TIMERS note explains quantized start offsets |
| Frame rate, headless (macOS) | 60 fps | 60 fps | 30 fps | Recorded as frameIntervalMs; LOW_FRAME_RATE (info below 50 fps, warning below 20 fps) |
scrollY precision | whole pixels | fractional | whole pixels | Scroll settles when the drift over the still period is below the threshold, not each frame's step |
| Keyframe offsets | as written | single precision | as written | Offsets rounded to 6 decimals (0.4000000059604645 → 0.4) |
| Start time vs. input event | ~25–45 ms later | at or before | ~2–4 ms later | Trigger latency is compared within one engine only |
| Interrupted transition timing | frame-quantized | frame-quantized | frame-quantized | The reversal's shortened duration depends on when the frame lands: compare within one engine |
cubic-bezier() near the endpoints | reference | same as Chromium | ~0.02 px apart | Below positionTolerancePx (2 px) |
Two problems these differences exposed were fixed for every engine: replay now seeks clones
in window-relative time rounded to 1 µs (Firefox resolved times that landed a hair below a
steps() boundary to the previous step), and scroll velocities are derived between samples
observed while moving, over at least 12 ms (two Lenis samples 1 ms apart on a 1 ms timer once
read as "78 px/ms", and the first frame of a 30 fps WebKit scroll was timed from a re-stamped
rest sample). The demo app's slower Lenis lerp now reads as a 30–36% lower peak in every
engine. Trigger latency is measured from the interaction that caused the motion, and a scroll
whose last still and first moved samples share a millisecond still starts from its rest position.
Unsupported and untested
- Safari, Chrome and Firefox as installed apps. WebKit is Playwright's WebKit build, close to
Safari but not Safari. Installed browsers can be used through
browser.channel(Chrome, Edge, and Playwright'smoz-firefoxchannels) orbrowser.executablePath; MotionPR's tests cover only Playwright's builds. - Other versions of an engine than the ones above are not tested.
- Windows is not covered by CI for any engine. Playwright's WebKit on Windows is the least mature build.
- Engines run one after another in a multi-browser run, not in parallel.
- Cross-browser pixel comparison is never definite: pixel, geometry, start-offset and scroll
differences between engines are
uncertainby design.
Engine setup and troubleshooting: Chromium, Firefox, WebKit.