prefers reduced motion os setting tells

prefers-reduced-motion and other OS-setting tells

Everyone spends their fingerprinting effort on the JavaScript surface: navigator, canvas, WebGL, audio. Meanwhile a stylesheet can read a cluster of your operating system's accessibility settings, with no script running at all, and report the answer back to a server before your first injected line executes.

prefers-reduced-motion is the best known of the group, but it travels with three siblings that behave the same way: prefers-contrast, forced-colors, and prefers-reduced-transparency. All four are pure CSS media features that mirror an OS accessibility setting, and all four are exfiltrable with a background-image branch and no JavaScript. This page covers what they expose, why the default value is the only safe one, how to check yours, and the one honest limit of fixing them.

The accessibility-media-feature cluster

These are not display-hardware features like resolution or color-gamut. They are user-preference features: each one reflects a toggle in the operating system's accessibility or personalisation settings, and the browser reports whatever the OS tells it.

Media feature	OS setting it mirrors	Default on a fresh Windows install
prefers-reduced-motion	Show animations / reduce motion	no-preference
prefers-contrast	High-contrast preference	no-preference
forced-colors	High-contrast mode active	none
prefers-reduced-transparency	Transparency effects off	no-preference

The pattern to notice: the overwhelming majority of real desktop users never touch any of these. They ship at their defaults and stay there. That is what makes a non-default value informative. A user who has turned on reduce-motion is rare, so reporting reduce narrows you to a small slice of the population before anything else is measured. The trap is not reporting an "unusual" value in the abstract. It is reporting a value that no untouched Windows install would report, which reads as a machine that was configured by hand.

Why no page-level layer can touch them

The mechanism is the same one that makes all of CSS media-query fingerprinting reach below the scripting layer. A stylesheet asks the question and encodes the answer as a network request:

@media (prefers-reduced-motion: reduce) {
  body { background-image: url("/p?rm=reduce"); }
}
@media (prefers-reduced-motion: no-preference) {
  body { background-image: url("/p?rm=none"); }
}
@media (forced-colors: active) {
  body { border-image: url("/p?fc=active"); }
}

Only the branch that matches loads its image, so the server learns the setting from which URL is requested. No matchMedia call, no property to override, no object to inspect. There is no JavaScript function for a page-level stealth patch to intercept, which is precisely why a page-level layer cannot fix this at all. The value has to be correct at the source, in the browser's own media-feature evaluation, or it is wrong on the first paint.

That is also why the values have to be baked into the engine rather than injected. By the time any script you add could run, the stylesheet has already fired its request.

What invisible_playwright reports here

The short version: this project ships these four features at their stock Windows defaults, so a fresh session reads exactly like an untouched Windows install and none of them narrows you to a configured minority. The launch is the ordinary two lines, and the browser you get is a real Playwright Browser:

from invisible_playwright import InvisiblePlaywright

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    values = page.evaluate("""() => ({
        motion:  matchMedia('(prefers-reduced-motion: reduce)').matches,
        contrast: matchMedia('(prefers-contrast: more)').matches,
        forced:  matchMedia('(forced-colors: active)').matches,
        transp:  matchMedia('(prefers-reduced-transparency: reduce)').matches,
    })""")
    print(values)   # {'motion': False, 'contrast': False, 'forced': False, 'transp': False}

All four come back False, which is the reading a default machine gives. The important part is not any single value. It is that this cluster agrees with the rest of the identity: it reports the same "plain, untouched Windows desktop" story that the fonts, the system-colour palette and prefers-color-scheme report. A machine that claims a default light theme and then reports high-contrast forced colours has answered the same question two ways, and a contradiction is worth more to a detector than any single odd value.

The matches above are stable for the seed. Pass the same seed on the next run and the cluster reads the same, which is the property that lets you replay a session rather than guess at it. If you need a specific field held constant while the rest stays seed-derived, that is what pinning a fingerprint field is for.

The consistency rule: agree with prefers-color-scheme

prefers-color-scheme is the fifth member of this family and the one people already know, because it is close to a coin flip across real users rather than a near-constant. It is also the anchor the other four have to agree with.

The rule is simple to state and easy to get wrong: the accessibility cluster and the colour-scheme preference and the system-colour palette all describe one machine, and they must tell one story. A dark-theme persona that still reports a light system palette is a contradiction. A machine reporting forced-colors: active while claiming an ordinary default theme is another. When these disagree, the disagreement is the signal, and it survives every JavaScript patch because none of it lives in JavaScript. Read the CSS system-colour palette section for how the OS theme colours cross-check against the same preference, and the color-gamut and HDR media queries for the display-hardware half of the same surface.

To check your own build against a stock browser, read both and diff them field by field rather than trusting a score:

[
  'prefers-reduced-motion: reduce',
  'prefers-contrast: more',
  'forced-colors: active',
  'prefers-reduced-transparency: reduce',
  'prefers-color-scheme: dark',
].forEach(q => console.log(q, matchMedia(`(${q})`).matches));

On a default Windows install every line is false except possibly the colour scheme. If any accessibility line comes back true and you did not intend it, that is the tell.

The honest limit

This cluster is an OS and configuration surface, and getting it right is exactly the kind of thing this project is built to do: match what a real Windows browser reports, field by field, so the fingerprint, the TLS handshake and the driver layer all read as a genuine Firefox driven by a real person. That is why sessions pass most detection checks. It is a demonstration, not a slogan: the four features above read as defaults because the engine ships them as defaults.

What it does not do, and cannot: the build ships stock defaults, and it cannot change your IP reputation, your per-account quotas, your rate limits, or your behaviour and timing. A perfect accessibility cluster on a datacenter address that a thousand other clients are using this minute still loses on the address. You supply the parts that live outside the browser: a clean residential exit, human pacing, sensible per-account volume. This page fixes one small OS-setting surface. It does not fix the session. If your fingerprint is clean and you are still blocked, the reason is usually one of the other three layers.

Conclusion

The accessibility cluster is a small surface with an outsized ability to embarrass a disguise, because it sits below every layer people normally work at and because its safe value is boring. Report the defaults a fresh Windows install reports, keep the whole cluster agreeing with prefers-color-scheme and the system palette, and it stops being a tell. Report a rare setting nobody turned on by hand, or contradict the colour scheme, and it becomes one of the cheapest signals a page can collect. invisible_playwright handles the source-level part; the network and the behaviour are still yours to get right.

Short answers to the questions that lead here

Can prefers-reduced-motion really be read without JavaScript? Yes. A stylesheet branches on it and loads a different background image per branch, so the server learns the value from which request arrives. There is no script to block.

What should a normal Windows machine report? no-preference for reduced motion, contrast and reduced transparency, and none for forced colours. Almost everyone leaves all four at those defaults.

Does invisible_playwright spoof these? It ships them at stock Windows defaults, so a default session reads like an untouched install and the cluster agrees with the colour scheme and the system palette rather than contradicting them.

Why is a non-default value a problem if it is a real setting? Because it is rare. A value only a small minority of real users report narrows you to that minority before any other signal is measured, and a value that does not match your claimed theme is a straight contradiction.

If I get these right, will I stop being blocked? Not on its own. This is one OS surface. It does not change your IP reputation, your rate limits, your account quotas or your behaviour, and any of those can block a browser whose fingerprint is perfect.

Is turning on Firefox's resistFingerprinting a shortcut here? No. That mode is usually the wrong move for automation; it changes and breaks more than this cluster and tends to create its own tells.

Sources

• The CSS media-feature specifications for the four preference features named above, and their documented default values.
• This project's own release gates, which compare each media feature against a stock Windows browser on the same machine rather than reading a verdict.

See also: what media queries reveal about a machine, the color-gamut and HDR half of the same surface, and why a clean fingerprint is only one of four layers.

---

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. The accessibility cluster is easy to forget precisely because its right value is the boring one.