Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarThe Decoder→ original

Claude Learned to Build Sites with Live Text Inside 3D Graphics in Chrome 157+

Until now, developers had to choose between two options: regular HTML stays clickable but distorts poorly, while canvas, the surface for 3D effects, turns text into a flat image. HTML-in-Canvas places live buttons and text inside the canvas, and Claude can build with it: the demo includes 15 examples, from a knitted pattern to a 3D cube. It works in Chrome and Edge from version 157, and the technology has to be named explicitly in the prompt. For promo pages, this means creative freedom without losing accessibility.

AI-processed from The Decoder; edited by Hamidun News
Claude Learned to Build Sites with Live Text Inside 3D Graphics in Chrome 157+
Source: The Decoder. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Decoder ↗
Editor-in-Chief
Key metric #1Chrome and Edge support it from version 157; Safari and Firefox do not support it yet
Key metric #2The feature is available in Claude desktop apps from version 5.11492
Key metric #3Google is developing HTML-in-Canvas as a web standard with accessibility built in

Until now, web developers faced an unpleasant choice. Regular HTML consists of text, buttons, and links that can be clicked and selected, but it is hard to distort: wrapping a page around a 3D cube is almost impossible. Canvas is a drawing surface on which the browser can render anything, from particles to 3D, but for the browser itself, such a canvas remains a flat picture. Now Anthropic has shown that Claude can use the HTML-in-Canvas technology: real page elements are placed inside the graphics and remain interactive. The demo includes 15 examples, and everything works in Chrome and Edge starting from version 157.

What the trade-off was and what is changing

Imagine a site menu. As regular HTML, you can select a word in it with the mouse and click an item, and a screen reader for visually impaired users will read it aloud. But it is difficult to twist such a menu into a spiral or stretch it over a volumetric shape.

If you draw the same menu on a canvas, it can take any shape, but the text turns into pixels. You cannot select a word, the link does not click, and a screen reader hears nothing.

HTML-in-Canvas combines these two approaches. Live page elements are embedded directly into the canvas, so the menu can be rotated, bent, and stretched, while the words in it can still be highlighted and copied. Google is developing this as a web standard that preserves both interactivity and accessibility.

What Anthropic showed

The capability was announced on X by Felix Rieseberg, who works on Claude at Anthropic. According to him, with this feature Claude can build "strange and beautiful websites," and the process reminds him of the Flash era, when the internet was full of experimental interactive pages.

The post comes with a demo page featuring 15 examples. Among them are sites rendered as knitted patterns, 3D cubes, and "exploded" layered diagrams, where page elements are pulled apart in space. In every case, the content remains real text.

According to The Decoder, this continues a trend: Claude is getting better at visual tasks. Recently, Claude Fable 5.2 painted photorealistic pictures stroke by stroke using Python code it wrote itself.

Limitations and what to do in practice

The technology is still young and not supported everywhere:

  • Chrome and Edge, starting from version 157;
  • Claude desktop apps, starting from version 5.11492;
  • Safari and Firefox do not support it yet.

There is also a caveat about usage: Claude does not apply this technique on its own. The request must explicitly ask for HTML-in-Canvas. Without such a hint, the model will most likely produce conventional markup.

What this means for developers and product managers: first, for experiments, promo pages, interactive stories, and creative portfolios, the new approach offers visual freedom without losing accessibility, which was nearly unattainable before. Second, for production with a mass audience, you need a fallback: Safari and Firefox users will not see the striking page as intended. A sensible approach is progressive enhancement: a baseline version in plain HTML, with the eye-catching layer added on top for browsers that support it.

Finally, this is a reminder that the value of AI assistants for front-end work is growing not only because they write code faster. They are starting to master new, still unsettled web APIs, and the more precisely you name the technology you want, the more you get from the model. For now, it is more of a playground than an industry standard, but it already shows how a return to creative web might look.

Story

Key Questions Answered

What the trade-off was and what is changing?

Imagine a site menu. As regular HTML, you can select a word in it with the mouse and click an item, and a screen reader for visually impaired users will read it aloud. But it is difficult to twist such a menu into a spiral or stretch it over a volumetric shape. If you draw the same menu on a canvas, it can take any shape, but the text turns into pixels. You cannot select a word, the link does not click, and a screen reader hears nothing. HTML-in-Canvas combines these two approaches.

What Anthropic showed?

The capability was announced on X by Felix Rieseberg, who works on Claude at Anthropic. According to him, with this feature Claude can build "strange and beautiful websites," and the process reminds him of the Flash era, when the internet was full of experimental interactive pages. The post comes with a demo page featuring 15 examples. Among them are sites rendered as knitted patterns, 3D cubes, and "exploded" layered diagrams, where page elements are pulled apart in space.

What are the key details on Limitations and what to do in practice?

The technology is still young and not supported everywhere: Chrome and Edge, starting from version 157; Claude desktop apps, starting from version 5.11492; Safari and Firefox do not support it yet. There is also a caveat about usage: Claude does not apply this technique on its own. The request must explicitly ask for HTML-in-Canvas. Without such a hint, the model will most likely produce conventional markup.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…