# Canvas Fingerprinting: What a Proxy Changes

> Canvas fingerprinting explained with a real test: one Chromium canvas hash, same directly and through residential and mobile proxies; headless gives another.

[Home](https://quanticdata.io/) / [Blog](https://quanticdata.io/blog/) / Canvas Fingerprinting: What a Proxy Changes

# Canvas fingerprinting: how it works, and the same hash through three proxies

Anti-bot · Nov 20, 2026 · 6 min read · By [Aldo Morese](https://quanticdata.io/about/), founder of QuanticData

![Canvas fingerprinting in a test: a script draws text and a coloured box on a hidden canvas, reads it back with toDataURL and hashes the pixels; the same Chromium produced hash 9a58f1e3 headless and 84a1c0e4 with a window, identical directly and through Residential Basic, Residential Premium and Mobile, while the IP changed on every path.](https://quanticdata.io/blog/canvas-fingerprinting/hero.webp?v=20261120h)

On this page

- [How canvas fingerprinting works](https://quanticdata.io/blog/canvas-fingerprinting/#how-canvas-fingerprinting-works)
- [Our test: two hashes, eight runs](https://quanticdata.io/blog/canvas-fingerprinting/#our-test-two-hashes-eight-runs)
- [What changes a canvas fingerprint, and what does not](https://quanticdata.io/blog/canvas-fingerprinting/#what-changes-a-canvas-fingerprint-and-what-does-not)
- [Why canvas fingerprinting matters for automation](https://quanticdata.io/blog/canvas-fingerprinting/#why-canvas-fingerprinting-matters-for-automation)
- [Test your own canvas fingerprint](https://quanticdata.io/blog/canvas-fingerprinting/#test-your-own-canvas-fingerprint)

Canvas fingerprinting is a tracking technique: a script draws text and shapes on an invisible HTML5 canvas, reads the pixels back and hashes them. Fonts, anti-aliasing, the graphics card and its driver make the result differ slightly between devices, and stay the same on one device. In our test the hash never changed through a proxy; switching the browser to headless did.

**Run it on the free credit.** Every QuanticData account gets $2 of free API usage per month.

[Start free — $2/month included](https://quanticdata.io/signup/?utm_source=blog&utm_medium=website&utm_campaign=blog-lead&utm_content=canvas-fingerprinting) · [Explore Web Scraping API](https://quanticdata.io/web-scraping-api/)

We ran it on 20 November 2026 on one Windows 11 PC with Chromium 149 driven by Playwright: headless and with a window, each directly and through QuanticData Residential Basic, Residential Premium and Mobile. Eight runs, the same drawing code each time.

## How canvas fingerprinting works

1. **Draw.** The script creates a canvas element that is never shown and draws on it: text in a given font, an emoji or two, overlapping shapes, a semi-transparent colour.
2. **Read back.** It exports the pixels with `toDataURL()` or `getImageData()`.
3. **Hash.** It reduces the image to a short value and sends it home with the rest of the fingerprint.

The drawing instructions are identical everywhere; the pixels are not. Font rendering, sub-pixel smoothing and the graphics pipeline round colours and edges slightly differently from one machine to the next, so the same few lines of code give a value that separates devices without a cookie. This is the probe we ran, the canvas half of it as it stood in our script:

```javascript
const c = document.createElement('canvas'); c.width = 280; c.height = 60;
const x = c.getContext('2d');
x.textBaseline = 'top'; x.font = '16px Arial'; x.fillStyle = '#f60'; x.fillRect(100, 1, 62, 20);
x.fillStyle = '#069'; x.fillText('QuanticData canvas test, 20 Nov', 2, 15);
x.fillStyle = 'rgba(102,204,0,0.7)'; x.fillText('QuanticData canvas test, 20 Nov', 4, 17);
const data = c.toDataURL();
let h = 0; for (let i = 0; i < data.length; i++) { h = ((h << 5) - h + data.charCodeAt(i)) | 0; }
(h >>> 0).toString(16)   // the canvas hash
```

Real fingerprinting scripts draw more and hash with a stronger function, but the mechanism is this one. WebGL fingerprinting is the 3D sibling: the script renders a scene and also reads the graphics card's vendor and renderer strings, which the `WEBGL_debug_renderer_info` extension exposes directly.

## Our test: two hashes, eight runs

| Chromium 149 mode | Path | Canvas hash | Data URL length | WebGL renderer |
| --- | --- | --- | --- | --- |
| Headless | Direct | `9a58f1e3` | 9,794 | ANGLE, SwiftShader (software) |
| Headless | Residential Basic, Premium, Mobile | `9a58f1e3` on all three | 9,794 | ANGLE, SwiftShader (software) |
| With a window | Direct | `84a1c0e4` | 9,746 | ANGLE, the PC's NVIDIA card |
| With a window | Residential Basic, Premium, Mobile | `84a1c0e4` on all three | 9,746 | ANGLE, the PC's NVIDIA card |

Each proxy path showed the test page a different US home or carrier IP, and not one pixel of the canvas changed. The data URL, the PNG the canvas was exported to, was 9,794 characters long in every headless run and 9,746 in every windowed run, so even its size gives the two modes away before any hash is computed. The WebGL vendor string told the same story: "Google Inc. (Google)" headless, "Google Inc. (NVIDIA)" with a window. A fingerprinting script reads those strings in one call, and a site that compares them with the User-Agent learns whether a browser that says it is desktop Chrome has a graphics card at all. Headless and windowed Chromium on the same PC disagreed, because headless mode drew with SwiftShader, a software renderer, instead of the graphics card: different rounding, different pixels, different hash. The WebGL strings said the same thing in words.

**Try it on your own targets.** Every QuanticData account gets $2 of free API usage per month.

[Start free — $2/month included](https://quanticdata.io/signup/?utm_source=blog&utm_medium=website&utm_campaign=blog-inline&utm_content=canvas-fingerprinting) · [Explore Web Scraping API](https://quanticdata.io/web-scraping-api/)

## What changes a canvas fingerprint, and what does not

| Change | Effect on the canvas hash |
| --- | --- |
| A new IP, through any proxy | None in our test: the canvas is drawn on your machine |
| Headless instead of a window (Chromium) | Changed in our test: software renderer instead of the GPU |
| Another graphics card or driver | Expected to change: it is the main source of the differences |
| Installed fonts, browser version | Can change it when the drawn text uses them |
| A browser that adds noise on purpose | Changes per site and per session, by design |

The last row is how privacy browsers fight canvas tracking. Brave, for example, describes adding small random changes to canvas and WebGL output that differ between sites and between sessions, so the hash can no longer link two visits; it also removes canvas and WebGL from third-party scripts by default. Extensions sold as "canvas fingerprinting protection" use the same two ideas: block the read-back, or perturb it.

## Why canvas fingerprinting matters for automation

For tracking, a canvas hash is a cookie that survives clearing cookies. For bot detection it is a consistency check. A headless browser that claims to be a desktop Chrome but draws with a software renderer, or a thousand "different" visitors that all draw the same pixels from a thousand IPs, are both easy to spot. That is the lesson of our eight runs: rotating the IP leaves the drawing untouched: the canvas is drawn on the machine, so the network does not enter into it. The other signals a site combines with it, from headers to [TLS fingerprinting](https://quanticdata.io/blog/tls-fingerprinting-ja3-ja4/), are in [how websites detect bots](https://quanticdata.io/blog/how-do-websites-detect-bots/); the full fingerprint surface is in [how browser fingerprinting works](https://quanticdata.io/blog/how-to-browser-fingerprint/).

If you automate a browser to collect public pages, run it the plain way, with a window where you can and a [Playwright with a proxy](https://quanticdata.io/blog/playwright-proxy/) setup for the network, and accept that it identifies itself as automated. Disguising a fingerprint to pass as a different person is not something we write recipes for. For public pages that need a real browser, see the [Web Unlocker](https://quanticdata.io/web-unlocker/). For your own browser runs, [residential proxies](https://quanticdata.io/residential-proxies/) and [mobile proxies](https://quanticdata.io/mobile-proxies/) cover the network side.

## Test your own canvas fingerprint

- Paste the probe above into the browser console on any page and note the value; reload and run it again: on an ordinary browser it stays the same.
- Run it in a private window, then in another browser on the same machine, and compare.
- Run it through a proxy or VPN: if your browser does not add noise, the value will not move.

Whether a site may fingerprint you, and what it must tell you, depends on where you are: see [is browser fingerprinting legal](https://quanticdata.io/blog/is-browser-fingerprinting-legal/) and [device fingerprinting and the law](https://quanticdata.io/blog/is-device-fingerprinting-legal/).

### Sources & further reading

- [Wikipedia: Canvas fingerprinting](https://en.wikipedia.org/wiki/Canvas_fingerprinting)
- [MDN: HTMLCanvasElement.toDataURL()](https://developer.mozilla.org/en-US/docs/Web/API/HTMLCanvasElement/toDataURL)
- [MDN: WEBGL_debug_renderer_info extension](https://developer.mozilla.org/en-US/docs/Web/API/WEBGL_debug_renderer_info)
- [Brave: Fingerprint randomization](https://brave.com/privacy-updates/3-fingerprint-randomization/)

## FAQ

Quick answers on canvas fingerprinting.

[Something else? Ask us](mailto:hello@quanticdata.io)

### What is canvas fingerprinting?

A tracking technique in which a script draws text and shapes on a hidden HTML5 canvas, reads the pixels back and hashes them. Small differences in fonts, smoothing and the graphics stack make the hash differ between devices and stay stable on one device.

### Does a proxy or VPN change my canvas fingerprint?

No. The canvas is drawn on your own machine. In our test Chromium gave the same hash directly and through Residential Basic, Residential Premium and Mobile, while the IP changed on every path.

### Is a canvas fingerprint unique?

Not on its own: identical hardware, drivers and browser versions can produce the same hash. Combined with other values such as screen size, fonts, time zone and the WebGL renderer, it narrows a visitor down much further.

### How do I test canvas fingerprinting?

Run a small script that draws text on a canvas, calls `toDataURL()` and hashes the result, like the probe in this article, or open a public canvas test page. Compare the values across browsers, modes and networks.

### What is canvas fingerprinting protection?

A browser setting or extension that blocks canvas read-back or adds small random changes to it, so the hash differs between sites or sessions. Brave documents both: noise per site and session, and no canvas for third-party scripts by default.

### Why is a headless browser canvas hash different?

Headless Chromium drew with SwiftShader, a software renderer, instead of the graphics card in our test, so the pixels and the hash differed from the same browser with a window: `9a58f1e3` against `84a1c0e4`.

## The network side of your browser runs

Residential and mobile proxies give your Playwright or Puppeteer runs home and carrier IPs in the country you choose.

[Start free — $2/month included](https://quanticdata.io/signup/?utm_source=blog&utm_medium=website&utm_campaign=blog-end&utm_content=canvas-fingerprinting) · [Explore Web Scraping API](https://quanticdata.io/web-scraping-api/)

## Related reading

- [How Do Websites Detect Bots? Signals Tested](https://quanticdata.io/blog/how-do-websites-detect-bots/): Anti-bot · Websites detect bots by stacking signals: the IP and its network, the headers, the TLS handshake, what JavaScript finds in the browser, and how the visitor behaves. We read those signals from six real clients, five of them also through residential and mobile proxies, to show what a proxy changes: the IP, not the client.
- [TLS Fingerprinting: JA3 and JA4 Tested](https://quanticdata.io/blog/tls-fingerprinting-ja3-ja4/): Anti-bot · A TLS fingerprint is a hash of the hello your client sends before any HTTP header. We measured JA3 and JA4 for curl, Python Requests, httpx and Chromium, directly and through three proxy networks, and Node directly. For the four clients we ran through the proxies, the IP changed every time and the JA4 never did.
- [Honeypot Traps in Web Scraping, Tested](https://quanticdata.io/blog/honeypot-traps-web-scraping/): Anti-bot · A honeypot trap is a link or form field that a person cannot see and a careless crawler follows. We built a test page with eight trap patterns, caught all eight with a visibility check in Playwright, then ran the check on four real sites: hundreds of hidden links, and the ones we sampled were ordinary menus.

---

Source: https://quanticdata.io/blog/canvas-fingerprinting/ · Site index for AI: https://quanticdata.io/llms.txt
