Skip to main content
Digital canvas displays interactive HTML elements, reviving Flash-like graphics. A user's hand hovers over the screen.

Editorial illustration for Claude Builds Interactive HTML Inside Canvas Graphics, Reviving Flash

Claude Builds Interactive HTML Inside Canvas Graphics,...

• 3 min read

Anthropic has added a feature to Claude that lets the model build web pages where text and buttons live inside canvas graphics but stay fully clickable and readable. That combination used to be impossible. Canvas elements render fast and bend into any shape a designer wants, but browsers treat them as flat images, no selectable text, no links a screen reader can parse.

Standard HTML does the opposite: accessible and interactive, but stiff when it comes to visual distortion. The new capability, called HTML-in-Canvas, lets Claude place real page elements directly into canvas-drawn scenes, so a button can sit on a warped 3D cube and still function like a button.

Felix Rieseberg, who works on Claude at Anthropic, posted about the update on X, framing it as a throwback to an older era of web design when pages chased visual tricks that today's accessibility-first standards generally avoid. Google is pushing HTML-in-Canvas toward becoming an official web standard, and browser support is already rolling out. The update follows a string of experiments showing Claude taking on more visual, less conventionally text-based tasks.

Claude can now embed text, buttons, and other web elements into complex graphics and 3D effects. Web pages have always faced a tradeoff. Standard HTML elements are clickable, selectable, and readable by screen readers, but they're hard to warp visually.

Why this matters

For developers who build interfaces, HTML-in-Canvas is a real technical tradeoff finally getting addressed instead of papered over. The old choice, accessible but visually limited HTML versus flexible but inert canvas, has shaped a generation of web design decisions. If Claude can genuinely keep buttons clickable and text readable by screen readers while drawing them inside warped 3D graphics, that's a legitimate shift in what's buildable without custom engineering for every project.

We'd want to see how this holds up outside a 15-example demo page before calling it solved, especially around accessibility tooling that depends on predictable DOM structure. The "Flash era" comparison is telling, Flash was beloved for creative freedom and loathed for breaking basic web standards like search indexing and link sharing. Founders eyeing flashier product pages and researchers studying AI-assisted coding should watch whether this pattern holds at scale, or whether it's another case of a flashy demo masking fragile output underneath.

The next thing worth checking: how Claude's canvas-embedded elements behave in actual screen reader software, not just visually.

LIVE02:41NVIDIA HPC SDK Playbook Brings GPU-Accelerated Compilers to DGX Spark