mirror of
https://github.com/feder-cr/invisible_playwright.git
synced 2026-06-07 08:35:12 +02:00
229 lines
13 KiB
Markdown
229 lines
13 KiB
Markdown
# invisible_playwright
|
|
|
|
[](https://github.com/feder-cr/invisible_playwright/actions/workflows/tests.yml)
|
|
[](LICENSE)
|
|
[](https://www.python.org/downloads/)
|
|
[](https://www.mozilla.org/firefox/)
|
|
[](https://github.com/feder-cr/invisible_playwright/releases)
|
|
[](https://github.com/feder-cr/invisible_playwright/stargazers)
|
|
[](https://github.com/feder-cr/invisible_firefox/releases/tag/usage-counter)
|
|
|
|
[](https://it.linkedin.com/in/federico-elia-5199951b6)
|
|
|
|
**Stealth Firefox that passes every bot detection test. Drop-in Playwright replacement, fingerprint patched at the C++ level, not a JavaScript shim.**
|
|
|
|
|
|
|
|
|
|
## Why it's powerful
|
|
|
|
|
|
**Most other anti-detect browsers patch Chromium at the JavaScript level** - they override `navigator`, `WebGLRenderingContext.getParameter`, canvas APIs, and so on via injected scripts. This has two fatal problems:
|
|
|
|
1. **JS patches are detectable.** Anti-bots enumerate native function `.toString()`, check descriptor configurability, compare property enumeration order, watch for prototype mutations. Every patch leaves a fingerprint of its own. CreepJS has an entire battery of "lies detectors" built around this.
|
|
2. **Chromium itself is now suspect.** Residential-proxy bot traffic is overwhelmingly Chromium-based, so detectors weight anything Chromium-shaped as risky by default. Chromium-based forks inherit Chrome's open-source layers (BoringSSL, Blink, V8, ANGLE) cleanly, but they still cannot fully match Chrome in practice: Chrome ships closed-source components on top (Widevine, proprietary codecs, Google Update / Safe Browsing endpoints) that flip detectable JS feature flags and network signals, and forks lag Chrome's release cadence by days to weeks, leaving telltale version-specific behaviours that detectors lock onto.
|
|
|
|
**invisible_playwright patches Firefox at the C++ level.** The spoofed values come back out through the normal Gecko paths - there is no JS shim, no override, no `Object.defineProperty`. **From the page's point of view, the browser is just telling the truth.** Anti-bot lie-detectors have nothing to latch onto.
|
|
|
|
invisible_playwright spoofs **all the layers that matter, together, coherently**: Navigator, screen, GPU/WebGL, Canvas, fonts, audio, WebRTC, timezone, DevTools detection, SOCKS5 auth, and the rest. See [feder-cr/invisible_firefox](https://github.com/feder-cr/invisible_firefox) for the full per-layer breakdown of which C++ files are patched and why.
|
|
|
|
Everything is driven by preferences - no hardcoded values in the binary. You change one pref, you change the spoofed value.
|
|
|
|
---
|
|
|
|
## How it compares
|
|
|
|
The closest peer in the source-level patching space is **Camoufox** (Firefox, open source): same approach as ours, but in a roughly year-long maintenance gap with its base Firefox several majors behind. **CloakBrowser** ships a similar pitch for Chromium, but its binary is **closed source** (the source-level patches are not published, you only get the compiled output), and it still hits the Chromium reCAPTCHA ceiling. The commercial anti-detect browsers (**Multilogin**, **GoLogin**, AdsPower, Dolphin, Kameleo) are paid SaaS that overlay JS-layer spoofing on a patched Chromium. Managed profiles are nice but raw detection bypass sits below both Camoufox and us.
|
|
|
|
| | invisible_playwright | Camoufox | CloakBrowser | Multilogin | GoLogin |
|
|
|---|---|---|---|---|---|
|
|
| Engine | Firefox 150 | Firefox (~1 year old base) | Chromium | Chromium fork | Chromium fork |
|
|
| Patch depth | C++ source | C++ source | C++ source (binary only) | JS overrides | JS overrides |
|
|
| Maintenance | Active (weekly) | Gap (~1 year) | Active | Active SaaS | Active SaaS |
|
|
| Open source | ✅ MIT | ✅ MPL | ❌ Closed source | ❌ Closed source | ❌ Closed source |
|
|
| `.toString()` clean | ✅ | ✅ | ✅ | ❌ Detectable shims | ❌ Detectable shims |
|
|
| Canvas / WebGL / Audio | ✅ C++ | ⚠️ Drift vs current FF | ✅ C++ | ⚠️ JS override | ⚠️ JS override |
|
|
| SOCKS5 auth | ✅ Patched | ❌ | ⚠️ Playwright proxy | ⚠️ Varies | ⚠️ Varies |
|
|
| **reCAPTCHA v3 score** | **0.90** | ~0.3-0.5 | ~0.3-0.5 | ~0.3-0.6 | ~0.3-0.6 |
|
|
| FP Pro - bot detected | ✅ Not detected | ⚠️ Sometimes | ⚠️ Sometimes | ❌ Detected | ❌ Detected |
|
|
| CreepJS lies | ✅ 0 | ⚠️ Increasing | ✅ 0 | ❌ Multiple | ❌ Multiple |
|
|
| Cost | Free | Free | Free | From $99/mo | From $49/mo |
|
|
|
|
---
|
|
|
|
## Install
|
|
|
|
```bash
|
|
pip install git+https://github.com/feder-cr/invisible_playwright.git
|
|
python -m invisible_playwright fetch # one-time ~100 MB download, SHA256-verified
|
|
```
|
|
|
|
Supported platforms: **Windows x86_64**, **Linux x86_64**.
|
|
|
|
---
|
|
|
|
## Usage
|
|
### Random fingerprint per session
|
|
**100% Playwright-compatible** - sync and async, all methods, zero API changes. If you already use Playwright, switching is two lines:
|
|
|
|
```diff
|
|
- from playwright.sync_api import sync_playwright
|
|
- with sync_playwright() as p:
|
|
- browser = p.firefox.launch()
|
|
+ from invisible_playwright import InvisiblePlaywright
|
|
+ with InvisiblePlaywright() as browser:
|
|
```
|
|
|
|
Every session gets a unique, coherent fingerprint drawn from real-world Firefox telemetry (GPU / audio / fonts / ~400 other fields) and Bezier-curve mouse motion baked into the browser itself.
|
|
|
|
**Sync**
|
|
```python
|
|
from invisible_playwright import InvisiblePlaywright
|
|
|
|
with InvisiblePlaywright(proxy={"server": "socks5://...", "username": "u", "password": "p"}) as browser:
|
|
page = browser.new_page()
|
|
page.goto("https://example.com")
|
|
page.click("#submit") # mouse arcs to the button on a Bezier curve
|
|
```
|
|
|
|
**Async**
|
|
```python
|
|
from invisible_playwright.async_api import InvisiblePlaywright
|
|
|
|
async with InvisiblePlaywright(proxy={"server": "socks5://...", "username": "u", "password": "p"}) as browser:
|
|
page = await browser.new_page()
|
|
await page.goto("https://example.com")
|
|
await page.click("#submit")
|
|
```
|
|
|
|
The `browser` object is a `playwright.sync_api.Browser` / `playwright.async_api.Browser` - every Playwright method works as-is.
|
|
|
|
---
|
|
|
|
### Random fingerprint per session
|
|
|
|
```python
|
|
from invisible_playwright import InvisiblePlaywright
|
|
|
|
with InvisiblePlaywright() as browser:
|
|
page = browser.new_page()
|
|
page.goto("https://creepjs-api.web.app")
|
|
```
|
|
|
|
Every call samples a new coherent profile. Log the seed to reproduce interesting runs:
|
|
|
|
```python
|
|
sf = InvisiblePlaywright()
|
|
with sf as browser:
|
|
print("seed =", sf.seed)
|
|
# ...
|
|
```
|
|
|
|
### Reproducible fingerprint
|
|
|
|
```python
|
|
with InvisiblePlaywright(seed=42) as browser:
|
|
... # same GPU, same canvas hash, same audio context, every run
|
|
```
|
|
|
|
### Proxies
|
|
|
|
```python
|
|
proxy = {
|
|
"server": "socks5://gate.example.com:1080",
|
|
"username": "user",
|
|
"password": "pass",
|
|
}
|
|
with InvisiblePlaywright(proxy=proxy) as browser:
|
|
...
|
|
```
|
|
|
|
Schemes supported: `socks5`, `socks4`, `http`, `https`. Auth works on all of them (SOCKS5 via patched `nsProtocolProxyService.cpp`, HTTP/HTTPS via Playwright). DNS is routed through the proxy by default, no local leak.
|
|
|
|
### Timezone
|
|
|
|
The browser timezone follows `timezone=`:
|
|
|
|
```python
|
|
# default: timezone is auto-derived from the egress IP (proxy egress if a
|
|
# proxy is set, otherwise the host's own public IP)
|
|
with InvisiblePlaywright(proxy=proxy) as browser:
|
|
...
|
|
|
|
# explicit IANA zone always wins — the only way to force a specific zone
|
|
with InvisiblePlaywright(proxy=proxy, timezone="America/New_York") as browser:
|
|
...
|
|
```
|
|
|
|
| `timezone=` | with proxy | without proxy |
|
|
|---|---|---|
|
|
| `""` (default) / `"auto"` | auto from proxy egress IP | auto from host public IP |
|
|
| `"Area/City"` | that zone | that zone |
|
|
|
|
The timezone always tracks the actual egress, so it can't disagree with the IP — a proxy in a different country paired with the host timezone is the classic `timezone_mismatch` signal. The egress IP is mapped to its IANA zone with an offline database ([`daijro/geoip-all-in-one`](https://github.com/daijro/geoip-all-in-one)), which auto-updates against its weekly rebuild and is cached locally (point `STEALTHFOX_GEOIP_MMDB` at your own `.mmdb` to skip the download). On failure: with a proxy the launch raises rather than silently using the host zone (pass an explicit `timezone=` to override); without a proxy it falls back to the host timezone so a transient lookup failure can't break the launch.
|
|
|
|
### Pinning specific fingerprint fields
|
|
|
|
By default everything comes from `seed`. To force specific values while the rest stays seed-derived:
|
|
|
|
```python
|
|
with InvisiblePlaywright(
|
|
seed=42,
|
|
pin={
|
|
"gpu.renderer": "ANGLE (NVIDIA, NVIDIA GeForce RTX 4090 Direct3D11)",
|
|
"gpu.vendor": "Google Inc. (NVIDIA)",
|
|
"screen.width": 2560,
|
|
"screen.height": 1440,
|
|
"hardware.concurrency": 16,
|
|
},
|
|
) as browser:
|
|
...
|
|
```
|
|
|
|
Full list of pinnable keys, how pinning interacts with the Bayesian sampler, and common patterns are in **[docs/pinning.md](docs/pinning.md)**.
|
|
|
|
---
|
|
|
|
## CLI
|
|
|
|
```bash
|
|
invisible_playwright fetch # download the binary if missing
|
|
invisible_playwright path # print the absolute path to the cached binary
|
|
invisible_playwright version # wrapper and binary versions
|
|
invisible_playwright clear-cache # remove all cached binaries
|
|
```
|
|
|
|
## Public API for downstream integrations
|
|
|
|
When you're building a third-party fetcher (a Crawlee `BrowserPool` subclass, a changedetection.io plugin, an agno toolkit, a Skyvern backend) and need to own the browser lifecycle yourself, use the public helpers instead of `InvisiblePlaywright`:
|
|
|
|
```python
|
|
from playwright.async_api import async_playwright
|
|
from invisible_playwright import ensure_binary, get_default_stealth_prefs
|
|
|
|
async with async_playwright() as p:
|
|
browser = await p.firefox.launch(
|
|
executable_path=str(ensure_binary()),
|
|
firefox_user_prefs=get_default_stealth_prefs(seed=42),
|
|
)
|
|
```
|
|
|
|
`get_default_stealth_prefs(seed, *, pin, locale, timezone, extra_prefs, humanize, virtual_display)` returns the same dict that `InvisiblePlaywright(seed=..., locale=..., ...)` would inject. Same deterministic seed semantics, same humanize toggle, same `extra_prefs` overlay. `ensure_binary()` downloads the patched Firefox on first call and returns its absolute path.
|
|
|
|
> Important: pass `headless=False` to `firefox.launch()` and manage display hiding yourself (Xvfb on Linux, hidden desktop on Windows). Passing `headless=True` directly puts Firefox in true headless mode and skips the real rendering pipeline, which breaks canvas / audio / WebGL fingerprint coherence. The `InvisiblePlaywright` context manager does this translation automatically; the public helpers leave it to the caller.
|
|
|
|
For everyday Python usage the `InvisiblePlaywright` context manager is still the recommended entry point.
|
|
|
|
## Related projects
|
|
|
|
invisible_playwright takes a different angle than the major Firefox-hardening projects but stands on their shoulders:
|
|
|
|
- **[arkenfox/user.js](https://github.com/arkenfox/user.js)** - the canonical Firefox configuration for privacy/security hardening via prefs. Reading arkenfox is how you understand which `user.js` knobs matter; invisible_playwright goes further by patching the C++ source where prefs alone are insufficient (Canvas noise, WebGL parameter overrides, font whitelisting, WebRTC IP swap, DevTools detection bypass).
|
|
- **[LibreWolf](https://librewolf.net)** - a Firefox fork bundled with sensible privacy defaults. Same audience, different distribution model: LibreWolf ships a configured Firefox binary, invisible_playwright ships source patches + a wrapper for automation.
|
|
- **[Camoufox](https://github.com/daijro/camoufox)** - the most well-known open-source anti-detect Firefox project. We share design goals on the fingerprint-spoofing side; the implementation approach differs (Camoufox patches a wider surface and ships its own fingerprint database, while invisible_playwright sticks closer to vanilla and drives spoofing from a Bayesian sampler).
|
|
|
|
---
|
|
|
|
## License
|
|
|
|
MIT - see [LICENSE](LICENSE). The patched Firefox binary is distributed under the MPL-2.0 (Firefox upstream license). The C++ patches against mozilla-central that produce that binary are at [feder-cr/invisible_firefox](https://github.com/feder-cr/invisible_firefox).
|