---
title: "Canvas Fingerprinting: How It Works and How to Control It"
description: "Canvas fingerprinting hashes how your browser draws hidden text and shapes. Learn to read your result and choose a stable, believable value per profile."
url: "https://donutbrowser.com/zh/canvas-fingerprinting/"
---

1. [首页](https://donutbrowser.com/zh/)

2. [博客](https://donutbrowser.com/zh/blog)

3. Canvas Fingerprinting: How It Works and How to Control It

# Canvas Fingerprinting: How It Works and How to Control It

October 18, 2026

canvas fingerprinting browser fingerprinting anti-detect browser

**[Canvas fingerprinting](https://en.wikipedia.org/wiki/Canvas_fingerprinting) is a technique where a script draws hidden text and shapes on an HTML5 canvas, reads the pixels back, and hashes them. That hash alone doesn’t identify a person; it’s [one signal among many](https://fingerprint.com/blog/canvas-fingerprinting/), useful mostly in combination with others. If you run several identities, aim for a separate, stable, believable canvas value per profile, not a blank or shared one.**

You keep profiles apart on purpose: client accounts, regional stores, QA runs, privacy. When two of them report the same canvas result, or one reports a blank canvas, sites get a reason to link or flag them.

After reading, you’ll be able to read your own result, pick a defense, and check that canvas agrees with the rest of the profile. The maker of Donut Browser wrote it by trade-off, so it stays useful with any browser.

> **Key Takeaways**

> - The prevalence figures from [Acar et al. (2014)](https://en.wikipedia.org/wiki/Canvas_fingerprinting) are history, not a measure of current adoption.

> - [DrawnApart (NDSS 2022)](https://arxiv.org/abs/2201.09956) reports GPU differences between individual units, so identical hardware models can still look different to a tracker.

> - A blocked, blank or per-page-random canvas stands out to the sites you visit.

> - Even good spoofing doesn’t guarantee that no account gets banned; behavior and platform rules still count.

## How does canvas fingerprinting work?

Canvas fingerprinting works in four steps: create a hidden canvas, draw on it, read the pixels back and hash them. Small differences in fonts, graphics drivers and rendering make the hash vary between devices.

![Image generated](https://api.birchseek.com/m/VRnwl_cGkcLzJVbEfxTx3L2PgOQIqsUO.webp)

Researchers from Princeton University and KU Leuven University described the technique in their 2014 paper [_The Web never forgets_](https://en.wikipedia.org/wiki/Canvas_fingerprinting) (Acar et al., ACM CCS). Their routine looks like this:

1. Create a canvas element. It’s usually never shown on screen.

2. Draw text and shapes with `getContext('2d')`.

3. Read the pixels back with `toDataURL()`, `toBlob()` or `getImageData()`.

4. Hash the pixels. The hash is a short stand-in for thousands of pixel values.

This snippet is an illustration, not a copy of any real tracker:

```
const canvas = document.createElement('canvas');
const ctx = canvas.getContext('2d');
ctx.font = '16px Arial';
ctx.fillText('Hello, canvas 123', 10, 30);
ctx.fillRect(10, 40, 80, 20);
const data = canvas.toDataURL(); // pixels encoded as a PNG string
const hash = await crypto.subtle.digest('SHA-256', new TextEncoder().encode(data));
```

The script is identical on every device. The output is not, because the browser does the drawing, and each machine draws slightly differently.

You can see your own hash and image in our [free canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint).

## Why does the same canvas code give different results on each device?

The same canvas code gives different results because the pixels come out of a stack of software and hardware layers, and each layer can differ from one device to the next.

All three read methods (`toDataURL`, `toBlob` and `getImageData`) return pixels, so any drawing difference changes the hash. [GoLogin’s guide](https://gologin.com/blog/what-is-canvas-fingerprinting/) says rendering behavior varies “across operating systems, fonts, graphics libraries, and hardware configurations”.

Here are the causes, from software to hardware:

1. **Installed fonts.** If the requested font is missing, the browser substitutes another, and every letter changes shape.

2. **Text shaping.** The text engine decides how letters are spaced and joined, and different engines decide differently.

3. **Antialiasing and subpixel choices.** Smoothing blends edge colors on curves, so small rule changes alter edge pixels.

4. **Browser and OS graphics stack.** The same browser version hands drawing to different system libraries on Windows, macOS and Linux.

5. **GPU and drivers.** [Fingerprint.com says](https://fingerprint.com/blog/canvas-fingerprinting/) canvas fingerprinting uses “subtle GPU and rendering differences that cause the same image to be drawn slightly differently across devices”.

The newest published step is [“DrawnApart: A Device Identification Technique based on Remote GPU Fingerprinting”](https://arxiv.org/abs/2201.09956) by Laor et al. (NDSS 2022). It goes below the model name and targets individual GPU units.

So a defense that only hides the GPU model name does not hide how that particular unit draws.

A canvas value only makes sense as part of a device story. A profile claiming a Windows laptop with one GPU needs canvas output that fits that claim. You can check yours with our [canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint).

## Is a canvas fingerprint unique to one person?

A canvas fingerprint isn’t unique to one person: many devices produce the same hash, and trackers get their value by combining it with other signals.

A canvas hash is closer to a shirt size than a name. Plenty of people wear a medium, but size plus height, shoe size and hair color narrows the crowd fast.

The other signals include the user agent, installed fonts, WebGL renderer, screen size and timezone, and our [browser fingerprinting guide](https://donutbrowser.com/browser-fingerprinting) covers each of them. [Fingerprint.com’s takeaways](https://fingerprint.com/blog/canvas-fingerprinting/) say it plainly: “Canvas signals remain useful when combined with other browser and device attributes rather than used in isolation.”

Stability depends on the browser. [GoLogin calls the output](https://gologin.com/blog/what-is-canvas-fingerprinting/) “a relatively stable signal.” That holds when a browser draws the same pixels every time.

Browser protections and extensions that randomize canvas output break that stability, as the next section shows. A randomized hash can change between sessions, so it can’t identify anyone alone. The GPU-level differences covered earlier only show up when nothing alters the output.

To check your own setup, run the [canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint) in each profile. Two profiles with the same hash are linked by that one signal. A blank value is an oddity of its own.

Aim for a stable, plausible value that matches each profile’s OS, GPU and fonts.

## Canvas blocking, random noise or consistent spoofing: which do you need?

Canvas blocking, random noise and consistent spoofing differ in what a site sees, and for people running several identities a stable, plausible value per profile is the one to aim for. This table judges them by logic, not by measured rates.

|  | Blocking | Random noise | Consistent spoofing |
| --- | --- | --- | --- |
| What the site sees | Blank or denied canvas | New output per page load or session | Same value on every visit |
| Breakage risk | Highest | Medium | Low |
| Detectability | High: an unusual result | Medium: repeat reads differ | Lower if it fits the profile |
| Identity | None | Per page or session | Stable per profile |
| Fits | Throwaway browsing | Casual privacy | Several identities |

[Fingerprint.com’s takeaways](https://fingerprint.com/blog/canvas-fingerprinting/) say “Browser protections increasingly limit or randomize canvas output, reducing reliability for standalone fingerprinting approaches.” That describes the tracker’s problem. Yours is different: a blank canvas stands out, and noise breaks continuity.

With per-page noise, a profile that loads a shop’s home page and then its checkout reports two different canvas hashes. The site may treat one visitor as two devices. A fixed value per profile gives it one device, which is what a real person has.

Where the defense lives matters too. The [Chrome Web Store listing for Canvas Blocker](https://chromewebstore.google.com/detail/canvas-blocker-fingerprin/nomnklagbgmgghhjidfhnoelnjfndfpd) says it “introduces slight changes to the data generated by the canvas element when specific methods are invoked.”

That is a page-script override: it covers the methods it wraps, and only those. Engine-level spoofing changes the output before page code reads it, so there is no wrapper for a script to find.

Whatever you pick, the value must fit the rest of the profile. A canvas result from a different OS or GPU than the user agent and WebGL vendor claim is itself a mismatch. Check a profile with our [canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint) before you rely on it.

If you want to look like everyone else instead, you need different tools, such as Tor Browser or Firefox’s resistFingerprinting. Check their current documentation for how they treat canvas.

- **Blocking** suits throwaway sessions where breakage doesn’t matter.

- **Random noise** suits casual privacy with no accounts to protect.

- **Consistent spoofing** suits agencies, sellers and QA work that needs each profile to stay the same device.

## How do you keep canvas consistent with the rest of a profile?

A canvas value stays consistent with a profile when the image could plausibly come from the exact device that profile claims to be. Treat it as one line in a checklist, not a standalone setting.

Canvas output carries traces of the operating system, fonts, graphics libraries and hardware that drew it. Each of those has to agree with what the profile says elsewhere.

Run this checklist for every profile:

- **User agent and platform.** If the profile says it’s Chrome on macOS, the canvas result should look like something a Mac produces, not a Windows machine.

- **Operating system.** Text drawn on a canvas depends on the fonts the system has. A Windows-only font set claimed on a macOS profile is a contradiction you created yourself.

- **GPU and WebGL vendor and renderer.** Graphics hardware shapes how shapes and gradients are rasterized. The GPU named in the WebGL data should fit the claimed OS and device.

- **Installed fonts.** The font list and the canvas text rendering should tell the same story. A short Linux-style font list next to a macOS canvas image doesn’t fit.

- **One stable value per profile.** Keep the same canvas result across sessions. A value that changes on every page load looks like noise, not like a device.

- **A different value for every other profile.** Two profiles that share one canvas hash can be linked to each other, which defeats the reason for keeping them apart.

A common failure: a profile claims macOS but returns a canvas image shaped by a Windows font stack. Each signal looks fine alone, yet together they describe no real machine.

The same logic applies to a timezone that doesn’t match the IP. The [proxies guide](https://donutbrowser.com/proxies-for-anti-detect-browsers/) covers the IP, timezone and language side of the profile.

This is a consistency check you run yourself, not a description of how any specific detector works. Open the profile, record the canvas hash, then compare it with the WebGL and browser fingerprint checks on the same profile. A single real device should be able to report all of it at once.

## How do you check your own canvas result and what should you change?

You can check your own canvas result in a few steps: read the hash, reload, compare across profiles, then match it to the device the profile claims to be.

1. **Open the test and note the hash.** Load the [free canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint/) in the profile you want to check. Copy the hash and look at the image. A blank image means canvas is blocked.

2. **Reload and compare.** Reload three times, then close and reopen the profile. You want the same hash every time. A hash that changes means noise is added on each page load.

3. **Open a second profile and compare.** Run the same test there. You want a different hash. An identical hash means both profiles tell sites the same story.

4. **Check the device story.** Open the [WebGL fingerprint test](https://donutbrowser.com/tools/webgl-fingerprint/) and the [browser fingerprint test](https://donutbrowser.com/tools/browser-fingerprint/). GPU, OS and user agent should describe one machine. A Windows user agent with an Apple GPU is a mismatch.

5. **Read the result.** Stable across reloads, different between profiles and plausible for the claimed device is a pass. Any one failure points to a fix in the table below.

The [Donut Browser canvas test](https://donutbrowser.com/tools/canvas-fingerprint/) draws the canvas image and shows the exact hash a tracker would take. [BrowserLeaks has a canvas page](https://browserleaks.com/canvas) too, and it lists image details such as file size and number of colors.

Run both in the same profile as a cross-check. Each page draws its own image, so the two hashes will likely differ. What matters is that each one stays stable on reload.

| Symptom | What it means | Fix |
| --- | --- | --- |
| Blank or blocked canvas | A visible choice that sites can notice | Move to a stable spoofed value |
| Hash changes on every reload | Per-page noise | Use a stable per-profile value |
| Two profiles share a hash | They share one fingerprint | Give each profile its own fingerprint |
| Canvas fits a different OS than the user agent | The profile tells two device stories | Regenerate the profile instead of patching one signal |

No single result proves a profile is safe, because canvas is one signal among many. Use the check to catch obvious contradictions before a site does.

## How does Donut Browser handle canvas fingerprinting?

Donut Browser sets canvas, like the rest of each profile’s fingerprint, inside its [Wayfern engine](https://donutbrowser.com/docs/wayfern/). In practice that means five things.

1. **Engine, not script.** Wayfern is a Chromium fork. Per the [Wayfern docs](https://donutbrowser.com/docs/wayfern), spoofing for canvas, WebGL, audio and more is set at the C++ level, not with injected page scripts. An extension like Canvas Blocker covers only the methods it wraps; the engine approach covers what the docs list. We don’t claim it’s undetectable.

2. **One device story per profile.** Canvas must agree with the user agent, OS, GPU, WebGL vendor and fonts. Each profile gets its own fingerprint and cookie store, as our [anti-detect browser explainer](https://donutbrowser.com/what-is-an-anti-detect-browser/) describes.

3. **Stable, not random.** Reload a profile and its canvas hash should stay the same. Automation that reuses the profile should see the same value too.

4. **Automation needs Pro.** The [Pro page](https://donutbrowser.com/pro/) describes launching a profile with one API call and attaching Playwright or Puppeteer over CDP, and [Donut’s pricing page](https://donutbrowser.com/pricing) lists that Browser Manipulation API on Pro. See our [automation use case](https://donutbrowser.com/use-cases/automation).

5. **Stated limits.** An anti-detect browser [can’t guarantee that accounts never get banned](https://donutbrowser.com/multiple-social-media-accounts-without-bans/). Behavior and platform rules still matter.

[The pricing page](https://donutbrowser.com/pricing) lists full fingerprint protection on the free plan, at the same quality as paid plans, and commercial use from Solo. The app [sends no telemetry](https://donutbrowser.com/no-telemetry/).

Run the [free canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint) on a profile before you rely on it.

## Frequently asked questions

### Can canvas fingerprinting identify me on its own?

**No, a canvas hash is one signal that many devices can share.** [Trackers combine it](https://fingerprint.com/blog/canvas-fingerprinting/) with user agent, fonts, WebGL, screen size and other signals, and the combination is what narrows a visitor down. A canvas value is evidence in a larger fingerprint, not a name tag.

### Do incognito mode or a VPN stop canvas fingerprinting?

**Neither one changes how your device draws the canvas.** [GoLogin’s guide](https://gologin.com/blog/what-is-canvas-fingerprinting/) explains that the output depends on your operating system, fonts, graphics libraries and hardware, and incognito and a VPN touch none of those. The hash usually stays the same across sites on the same device, unless your browser randomizes it.

### Is blocking canvas better than spoofing it?

**It depends on your goal.** Blocking and random noise are visible choices: a blank canvas or a value that changes on every read looks unusual. A stable spoofed value suits separate identities, as the comparison table above lays out. You can [check what a profile reports](https://donutbrowser.com/tools/canvas-fingerprint) before you rely on it.

### Does canvas spoofing guarantee my accounts won’t be banned?

**No, it only removes one signal that links accounts.** Your IP, cookies, behavior and each platform’s rules still count. A profile with a plausible canvas value and spammy activity will still get flagged.

## What should you do next?

Make canvas checks routine instead of a one-time fix.

- **Today:** Run the [canvas fingerprint test](https://donutbrowser.com/tools/canvas-fingerprint/) in each profile and note the hash. Reload and compare.

- **This week:** Remove any blank, shared or per-page-random canvas setup. Then check that canvas agrees with the user agent, OS and WebGL each profile claims.

- **Ongoing:** Recheck after browser updates and after you create new profiles.

To try engine-level spoofing, download [Donut Browser](https://donutbrowser.com), which offers unlimited local browser profiles, zero telemetry and downloads for macOS, Windows and Linux, [with no account needed](https://donutbrowser.com/pricing/).

Open one profile and test it with the canvas tool. If the result doesn’t fit the device you want, you’ve spent nothing. The Pro plan matters only if you need CDP automation with Playwright, Puppeteer or Selenium.

## Sources

[What is Canvas Fingerprinting? Myths, Tests & Real…](https://gologin.com/blog/what-is-canvas-fingerprinting/) · [Canvas Fingerprinting: What It Is and How It Works](https://fingerprint.com/blog/canvas-fingerprinting/) · [Canvas fingerprinting - Wikipedia](https://en.wikipedia.org/wiki/Canvas_fingerprinting) · [DrawnApart: A Device Identification Technique based on…](https://arxiv.org/abs/2201.09956) · [Canvas Blocker - Fingerprint Protect - Chrome Web Store](https://chromewebstore.google.com/detail/canvas-blocker-fingerprin/nomnklagbgmgghhjidfhnoelnjfndfpd) · [Canvas Fingerprint Test — Donut Browser](https://donutbrowser.com/tools/canvas-fingerprint/) · [Canvas Fingerprinting - BrowserLeaks](https://browserleaks.com/canvas) · [Free AdsPower Alternative, Open Source | Donut Browser](https://donutbrowser.com/alternatives/adspower/). Every other source is linked where the article cites it.

## 在 Google 中更常看到我们

只需点击一次，即可在您的 Google 账号中将 Donut Browser 设为首选来源，我们的文章将更有可能出现在您的热门新闻、AI 模式和 AI 概览中。

在 Google 中设为首选来源

## On this page

- How does canvas fingerprinting work?

- Why does the same canvas code give different results on each device?

- Is a canvas fingerprint unique to one person?

- Canvas blocking, random noise or consistent spoofing: which do you need?

- How do you keep canvas consistent with the rest of a profile?

- How do you check your own canvas result and what should you change?

- How does Donut Browser handle canvas fingerprinting?

- Frequently asked questions

- Can canvas fingerprinting identify me on its own?

- Do incognito mode or a VPN stop canvas fingerprinting?

- Is blocking canvas better than spoofing it?

- Does canvas spoofing guarantee my accounts won’t be banned?

- What should you do next?

- Sources
