Graphics Fingerprint

doc

Analyze Canvas and WebGL rendering to understand GPU fingerprinting. Detect hash stability issues and reduce tracking surface with actionable diagnostics.

Two machines with the same browser, the same version and the same OS can still render the same page differently. Graphics fingerprinting measures that difference. It is precise, it is stable across sessions, and it cannot be fully removed without degrading the rendering that users actually need.

Where the difference comes from

Rendering is not a pure function of the page. It depends on the pipeline below it:

  • GPU and driver. The graphics hardware and its driver determine which extensions exist, which precision modes are supported, and how antialiasing is implemented.
  • Font rendering. Hinting and rasterisation differ between systems and between font versions. Text drawn with the same CSS produces different pixels.
  • Colour management. Display profiles, gamma handling and wide-gamut support change the values written into a canvas.
  • Compositing. Sub-pixel positioning, layer blending and image resampling vary by platform.
  • Implementation details. Even within one browser, platform-specific paths exist for image decoding and text layout.

None of these are visible to the user. All of them are observable to code running in the page.

What gets measured

SurfaceWhat it revealsStability
Canvas text and shapesAntialiasing, font rasterisation, rounding behaviourHigh across sessions, changes with OS or font updates
WebGL renderer stringGPU vendor and driver family, sometimes exposed in masked formFixed per device
WebGL parameter setLimits, precision, supported extensionsFixed per device and driver
Shader output hashActual computed pixels from a fixed shaderHigh; sensitive to driver changes
Image decode resultDecoder and colour pathModerate

The canvas hash is the familiar one. A page draws text and shapes, reads the pixel data back, and hashes it. The hash is identical on repeat visits from the same machine and different on another machine with different hardware or fonts.

The double-edged property: stability

A graphics fingerprint is useful for two opposite purposes, and this is why it is hard to regulate:

  1. Fraud detection. A login from a machine whose canvas hash is identical to an earlier session is more likely the same device, even from a new IP. Banks and payment processors rely on this.
  2. Tracking without consent. The same stability lets a site recognise a returning visitor without any cookie, and without a consent prompt, because no data is "stored" in the usual sense.

The measurement is the same in both cases. What differs is the intent and the disclosure. A check page can show you the value; it cannot tell you what other sites do with theirs.

Why it cannot be fully removed

Proposals to randomise graphics output run into a simple constraint: the output has to be correct. Blurring a canvas, adding noise, or forcing software rendering changes what users see and how fast it renders.

  • Noise injection is detectable in aggregate. Repeated samples converge on a stable average, and statistical analysis recovers the underlying signal.
  • Software rendering makes the fingerprint uniform — and also makes the browser slow, and creates a new, very recognisable class.
  • Blocking WebGL entirely breaks legitimate applications, and the absence of WebGL is itself a distinguishing property.

The realistic outcome is not anonymity. It is a smaller but still distinctive set.

What to look at in a result

  1. Whether the values are exposed at all. Some browsers mask the GPU string; some do not. Masked is better than raw, but masking is not removal.
  2. Whether the hash is stable across two checks. A hash that changes between checks is less useful for tracking — and also less useful for fraud detection.
  3. Which surface dominates. Often it is the font set, not the GPU. A distinctive font list matters more than an unusual renderer name.
  4. Whether a change actually helps. New hardware changes the fingerprint. It does not hide it.

Common misreads

  • "A canvas hash identifies me." It identifies a rendering pipeline. Millions of machines share one.
  • "Blocking WebGL makes me anonymous." It changes one property and adds a conspicuous one: WebGL disabled.
  • "The hash is random noise." It is a deterministic function of hardware and software. It repeats exactly.
  • "A different hash means a different person." A driver update or an OS patch changes it legitimately.

FAQ

Is graphics fingerprinting the same as canvas fingerprinting?

Canvas is one surface. WebGL parameters, shader output and image decoding are others. The general term covers all of them.

Does a VPN hide it?

No. A tunnel changes the network path. Rendering happens locally and is unchanged.

Can two identical laptops produce the same hash?

Often yes, if hardware, driver and font versions match. That is why the hash narrows a population rather than selecting a person.

Should I disable WebGL for privacy?

Only if you do not need it. Disabling removes a capability and adds a distinctive configuration, which is a trade, not a win.