Claude now builds websites inside canvas graphics, and the HTML stays clickable
Anthropic's Claude can now build web pages with HTML-in-Canvas, a new browser technology. Text, buttons and links sit inside warped 3D graphics but you can still click and select them. One Anthropic developer says it reminds him of the Flash era. For now it only works in Chrome, Edge and the Claude desktop apps.

What's going on
Anthropic's AI assistant Claude has a new and rather unusual way to design web pages. Using a browser technology called HTML-in-Canvas, Claude can build real page elements like text, buttons and links into elaborate graphics and 3D effects. The interesting part is that those elements keep working. You can select and copy text, click buttons and follow links, even when the whole page is shown as a spinning cube or a knitting pattern.
Felix Rieseberg, who works on Claude at Anthropic, brought the feature to wider attention. On X he wrote that Claude can now use it to build "bizarre and beautiful pages" and said it reminded him of the Flash era. Outlets including The Decoder have covered it. The feature is available in Claude, but you need a supported browser: Chrome or Edge from version 157, or Claude's own desktop apps from version 5.11492.
The old trade-off: clickable or pretty
To see why this matters, it helps to look at how web pages normally work. Roughly speaking, there are two worlds:
- Regular HTML is the language web pages are made of. Text, headings, buttons and links are "real" building blocks. The browser knows what a link is, you can select text, search engines can read it, and screen readers (software that reads content aloud to blind and visually impaired people) can speak it. The downside is that you can only bend these elements so far visually. Folding a paragraph like fabric or wrapping it around a curved 3D surface is close to impossible with plain HTML.
- Canvas is a kind of digital drawing surface in the browser that a program paints on pixel by pixel. Almost anything is possible here: games, 3D scenes, particle effects. But to the browser, the result is just a flat image. Text in a canvas isn't text anymore, only painted pixels. You can't select it, copy it or click links in it, and screen readers see nothing.
So until now, if you wanted to build something visually spectacular, you had to choose between flashy effects and usability plus accessibility. Developers often worked around this with elaborate hacks, such as layering invisible HTML elements over a graphic.
What HTML-in-Canvas changes
HTML-in-Canvas gets rid of that trade-off. It places real page elements inside a canvas graphic without stripping away their properties. According to The Decoder, Google is developing the technology as a web standard meant to keep both interactivity and accessibility intact. A paragraph can be distorted, rotated or dropped into a 3D scene and still be a paragraph the browser recognizes as one.
That's exactly what Claude now takes advantage of. A demo page from Anthropic, or more precisely Rieseberg, shows 15 examples, including:
- web pages rendered as a knitting pattern,
- pages laid out on 3D cubes,
- and "exploded" layer views, where a page's individual layers are pulled apart and float in space.
In every case, the content stays real text you can select and copy. The source doesn't give the demo page's address.
The Flash comparison
Rieseberg's Flash comparison isn't random. In the 2000s, Adobe Flash was the go-to technology for playful, animated websites, intros and browser games. Many of those sites were creative and experimental, but they shared the weaknesses of canvas graphics: content was often invisible to search engines and screen readers, text frequently couldn't be copied, and you needed a separate plug-in. Flash was eventually discontinued.
HTML-in-Canvas promises to bring back some of that era's creative freedom, this time built on open web technology and real HTML instead of a closed plug-in ecosystem. And because an AI can generate these pages on request, the barrier to entry drops even further. Work that used to need a specialized dev team is now something you can experiment with by describing it in text.
How to use it
According to The Decoder, a few practical points matter:
| Item | Details |
|---|---|
| Browser | Chrome and Edge from version 157 |
| Claude desktop app | from version 5.11492 |
| Not supported | Safari, Firefox (so far) |
| Activation | You must explicitly ask for HTML-in-Canvas in your prompt |
That last point is key, because Claude won't reach for the technology on its own. If you want it, ask for it explicitly in your prompt (the text you give the AI), for example: "Build me a portfolio page using HTML-in-Canvas where the sections flip like the pages of a book."
If you're on Safari or Firefox, you won't see the effects for now. The same goes for anyone visiting a page built this way, which matters if you actually plan to publish one.
Claude keeps getting more visual
The Decoder says the new feature is part of a series of visual experiments. Just recently, Claude Fable 5.2 painted photorealistic artworks brushstroke by brushstroke using Python code it wrote itself. Anthropic is positioning Claude not just as a coding and writing assistant but more and more as a tool for design work too.
Reports like one from Geeky Gadgets point the same way. It describes how you can use Claude to generate entire websites, complete with animations and interactive 3D elements, from a text description. That report covers website building with Claude in general, though, not HTML-in-Canvas specifically.
The bigger picture
The feature is mainly exciting for web designers, creatives and anyone who likes to experiment. It shows how fast AI assistants pick up new web standards: Claude is using a technology that's only just arriving in browsers. That's appealing, but it has limits. Until Safari and Firefox catch up, plenty of people won't see the effects, especially on iPhones and Macs, where Safari is common. That's a problem for serious public websites, less so for experiments, demos and art projects.
On the plus side, the approach pairs spectacle with accessibility. Unlike Flash sites or classic canvas graphics, the content stays readable and usable. The source doesn't show in detail whether that works equally well on every generated page, for example with screen readers. It's also unclear how Google will push the standard forward and whether other browser makers will adopt it. Until then, HTML-in-Canvas in Claude is mostly a playground for unusual web interfaces that would have taken a lot of work to build just a few years ago.
Sources
- The Decoder (EN) the-decoder.com
- geeky-gadgets.com geeky-gadgets.com
- Reddit rising: KI-Szene claude.ai
- Merkliste: diariobitcoin.com diariobitcoin.com
- Merkliste: techtimes.com techtimes.com
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.