photomopOpen the studio

AI and on-device processing

WebGPU

Also called gpu in browser or web gpu api. Here is what it means, when it changes what you export, and what people get wrong.

AI and on-device processingRuns on your deviceUpdated 10 August 2026

The short answer

WebGPU is a W3C standard that gives web pages direct access to modern graphics hardware, for both rendering and general-purpose compute. It succeeds WebGL, and for image work its value is compute shaders: neural network maths that would crawl on the CPU runs in parallel on the GPU.

Updated

The short version

WebGPU at a glance

Quick facts about WebGPU: standardised by, replaces, key feature for ai, shipping by default in, firefox, fallback
Quick factDetail
Standardised byW3C
ReplacesWebGL for new work
Key feature for AICompute shaders
Shipping by default inCurrent Chrome, Edge and Safari
FirefoxRolling out by platform
FallbackWebAssembly on the CPU

Slide the table sideways to see every column.

What does WebGPU do that WebGL could not?

WebGL was designed around drawing triangles. You could coax it into general computation by pretending your data was a texture and your maths was a fragment shader, and people did, but it was awkward and left performance on the table.

WebGPU treats compute shaders as a first-class feature. You can dispatch arbitrary parallel work over buffers of numbers with no rendering pretence at all, which is exactly the shape of neural network inference. It also maps far more closely onto the modern native APIs underneath it, with explicit pipelines, command encoders and resource binding, so less time is lost to driver guesswork.

Which browsers can you count on today?

WebGPU ships by default in current versions of Chrome, Edge and Safari, with Safari support arriving in Safari 26 in 2025. Firefox support has been rolling out platform by platform rather than everywhere at once, so treat it as present on some builds and absent on others.

Browser version is not the whole story either. Availability also depends on the machine: the GPU, its drivers, and whether the browser has blocklisted that combination. The only reliable test is a runtime one. Check for navigator.gpu, then request an adapter and a device, and treat a failure at any of those points as WebGPU not being available rather than as an error to report.

What happens when WebGPU is not available?

You fall back to the CPU, and in a browser that means WebAssembly. The same model and the same operators run through a compiled runtime instead of the graphics hardware. Results should match closely, though minor numeric differences between backends are normal and occasionally visible at a mask boundary.

The difference you will notice is time. A GPU has thousands of small cores working simultaneously, which suits the large matrix multiplications inside a network far better than a handful of CPU cores. Plan for both paths: keep the WebAssembly route working, and size your default input resolution so the fallback is still tolerable rather than only usable on the fast path.

What people get wrong about webgpu

Each one is a real failure mode, not a style preference.

  1. Checking the browser name or version instead of requesting an adapter, which misses machines where the GPU or driver is blocklisted.

  2. Shipping a GPU-only path with no WebAssembly fallback, so the tool simply fails on hardware it cannot use.

  3. Assuming a supported browser guarantees fast hardware, when integrated graphics on a thin laptop can be slower than expected.

  4. Reading pixels back from the GPU more often than necessary, since the transfer can cost more than the computation it interrupts.

Questions people ask

WebGPU, answered

The follow-up questions people search for once they have the definition.

Do I need WebGPU to remove a background in a browser?

No. A cut-out sized model runs on the WebAssembly CPU path as well, just more slowly. WebGPU turns a several-second wait into something closer to immediate, and it becomes far more important for heavier work such as upscaling or processing a batch of images.

Is WebGPU a security risk?

It exposes the GPU through a deliberately constrained interface, with bounds-checked memory access and validated shaders rather than raw driver access. The design goal was to give web pages hardware speed without giving them the freedom that native GPU code has, and browsers additionally blocklist drivers known to be unsafe.

Why does my page say WebGPU is unavailable in a supported browser?

The most common causes are an unsupported or blocklisted GPU driver, a virtual machine or remote session without hardware acceleration, or a browser flag or policy that disables it. It is a device-level condition, so the fallback path has to handle it rather than the page reporting a failure.

Does WebGPU replace Canvas for image editing?

Not entirely. Canvas remains the practical way to decode, draw and export images, and most pipelines use it for input and output. WebGPU takes over the heavy per-pixel maths in the middle, then results are written back for display or encoding.

Sources

Where the facts on this page come from. Every link opens the specification, standard or documentation the claim was read at.

Now do it to a photo

Photomop is a photo studio that runs on your own device. Resize, crop, compress, convert, batch edit, remove a background and change one all work in the browser tab you are reading this in, at full resolution and with no watermark. The on-device page shows you how to check that the photo stays put.