vs camoufox

Playwright stealth vs Camoufox: two patched Firefoxes

invisible_playwright and Camoufox both compile Firefox from source and set the fingerprint in C++, so neither is stealthier in the abstract. The real difference is how each produces the identity: Camoufox rotates a fresh, plausible device per session, while invisible_playwright derives every field deterministically from one seed, so a failing run can be replayed exactly. Camoufox currently covers more surfaces, notably geolocation and Intl.

Most comparisons in this space are between things that are not comparable: a page-level script against a patched binary, or a Chromium tool against a Firefox one. This one is not. Camoufox and this project take the same approach, on the same engine, for the same reason.

So the useful question is not which is stealthier in the abstract. It is what each one does differently, where each is ahead, and which properties matter for your job.

This page is written from Camoufox's own documentation and from ours. Where I could not verify something about their implementation, I say so rather than guessing.

invisible_playwright vs Camoufox at a glance

Both patch the same engine; the table below is where they diverge. Each row is expanded in its own section below.

Property	invisible_playwright	Camoufox
Approach	Compiles Firefox from source, patches the C++	Compiles Firefox from source, patches the C++
Identity model	Derived deterministically from one seed	Rotated per session from a real-world distribution
Optimises for	Reproducibility (replay a failing run)	Variety (a fresh device each time)
Replay a failing run exactly	Yes, rerun the seed	Not the design goal
Geolocation spoofing	No	Yes (documented)
Non-English Intl correctness	Known weakness, not closed	Documented Intl spoofing
Fonts	Real Windows fonts bundled in the binary, host-independent	Spoofing and masking of the host's fonts
Works with existing Playwright code	Yes	Yes

What the two have in common

Enough that the shared part is worth stating once.

Both compile Firefox from source and modify the C++, rather than injecting JavaScript into the page. That single decision is what separates either of them from the page-level tools, and the reasons are the same in both cases: an override is a function whose source can be printed, a worker asks the browser rather than your patch, and an injected script has to win a race it does not always win.

Both work with existing Playwright code, so switching is a launch change rather than a rewrite.

Both cover broadly the same surfaces: navigator, screen and viewport, WebGL parameters and extensions, canvas, fonts, audio, WebRTC, timezone and locale.

Both are Firefox, which means both carry the same market-share cost and both avoid the Chromium-is-not-Chrome problem entirely.

If you are choosing between these two and a page-level tool, that choice is already made. The rest of this page is about a narrower question.

Where the identity comes from

This is the real architectural difference and everything else follows from it.

Camoufox rotates device characteristics using a generator that matches real-world traffic distributions, per its own description. The emphasis is on producing a fresh, plausible device each time, drawn from how devices are actually distributed.

This project derives every field from a single integer. One seed produces one machine: the same GPU, the same canvas hash, the same audio profile, the same fonts, every run, forever. The sampler is conditioned across dimensions so the values agree with each other rather than being drawn independently.

Neither is better in the abstract. They optimise for different things:

• Rotation optimises for variety. A fresh device per session, and no work needed to get one.
• Derivation optimises for reproducibility. A failing run can be replayed exactly, which is the difference between bisecting a problem and guessing at it.

That second property is why this project is built that way, and it is worth being concrete about what it buys. When a run fails, you rerun the seed and get the identical browser. When you need a hundred identities, you use a hundred integers and each one is a stable, returnable machine. When you pair a profile with an identity, the seed and the directory stay married.

If your work is one-shot and volume matters more than replay, that property is worth nothing to you and rotation is the simpler answer.

Where Camoufox covers more

Being straight about this, because a comparison that only lists your own advantages is marketing.

Camoufox's documented surface list includes geolocation spoofing and explicit Intl spoofing. This project does not do geolocation, and its locale handling has a known weakness on the Intl side that we have not closed.

So on breadth of surface, on the evidence of what each documents, Camoufox covers more. If your target reads geolocation, that difference decides it.

It also ships its own fingerprint generation as a first-class, separately usable component, which is a real engineering asset and a reason people build on it.

Where the approaches differ on fonts

Both do something about fonts, and the mechanisms are not the same.

Camoufox describes font spoofing and anti-fingerprinting, which is a filtering and masking approach applied to what the host has.

This project ships real Windows fonts inside the binary and reads only from that bundle, so the font list and the rendered glyph metrics are the same on a Linux container and on a Windows desktop. It is a heavier binary in exchange for a font surface that does not depend on the host at all.

Why fonts are a comparison rather than a count explains why this surface is worth the weight either way.

Where I could not verify a difference

Two areas where our documentation says something and I could not confirm the equivalent on their side, so I am not claiming an advantage.

Pointer event fields. Both projects advertise human-like mouse movement. This project sets pointerType, pressure and the trusted flag in the browser's own input handling, so the events carry what a real device produces rather than what a synthesiser guessed, which is a different thing from drawing a good curve. Whether Camoufox does the same at that level is not something their front page states and I have not read their source for it.

The automation layer's own artefacts. We found and fixed four leaks in the driver rather than the engine: JIT-disabling timing, driver frames in stack traces, evaluated code labelled as debugger evaluation, and a helper running in the page realm before the page. Those are here. Any Playwright-driven Firefox has the same layer and therefore the same starting position, and I do not know which of them they have addressed.

If you are choosing seriously, those are the two things worth testing yourself rather than reading about. The method for testing them is the same for both.

How to actually choose

Working from the above rather than from claims:

• Need geolocation, or non-English Intl correctness? Camoufox, today.
• Need to replay a failing run exactly? A seeded derivation gives you that by construction.
• Need a font surface that is identical on a container and a desktop? The bundled approach does that at the cost of binary size.
• Need Chromium? Neither. Both are Firefox and the engine question is separate.
• Undecided? Run both against your actual target, ten times each, and compare the reports rather than the verdicts. Two hours of that beats any comparison page, including this one.

Conclusion

These two tools agree about the hard part. Patching the engine rather than the page is the expensive, correct decision, and both made it.

What is left is a genuine difference in philosophy: rotate a plausible device, or derive one reproducibly from a number. Plus a real difference in coverage, currently in their favour on geolocation and Intl, and a real difference in how fonts are handled.

Pick on the property your work needs. And whichever you pick, the machine you run it on still answers for its own GPU, and neither project changes that.

Short answers to the questions that lead here

Is Camoufox or invisible_playwright better? They take the same approach on the same engine. Camoufox currently covers more surfaces; this project derives everything from a seed so a run can be replayed exactly.

Do both patch Firefox at the C++ level? Yes. That is the shared decision and the thing that separates both from page-level tools.

Can I use my existing Playwright code with either? Yes, both are launch-level changes.

Which has better fingerprints? They are produced differently rather than better: one rotates from a real-world distribution, one derives deterministically from a seed with the dimensions conditioned on each other.

Does either fix my container's missing GPU? No. Nothing at the browser level does.

Should I run both? Never at the same time. Two components deciding the same values contradict each other, which is worse than either alone.

See also: three ways to make Playwright undetected, for why both of these sit at the same level, Firefox or Chromium for anti-detect, for the cost they share, and how to test whether your browser is detected, for how to settle this yourself.

Sources

• Camoufox's own documentation for its stated mechanism and surface list, read 2026-07-28.
• MDN's Intl reference and the MDN Geolocation API overview, for what each documented surface actually covers on the web-platform side.
• MDN's PointerEvent.pointerType, PointerEvent.pressure and Event.isTrusted references, for what the input event fields this project sets are specified to carry.
• This project's patch catalogue for the seeded sampler, the bundled font architecture, the input event fields and the four automation-layer fixes.
• Where a claim about their implementation could not be checked against their source, the page says so instead of asserting it.

---

From the notes of invisible_playwright, a Firefox patched at the C++ level. Camoufox got here first and the section where it covers more is the one we would want a comparison of ours to include.