Web design often asks creators to choose between visual freedom and usable page elements. Claude’s new HTML-in-Canvas capability aims to bring both together: it can put real text, buttons and other web elements inside intricate graphics while keeping them interactive.
Combining canvas visuals with real page elements
Standard HTML elements offer familiar functions: people can click them, select text and use screen readers to read page content. But these elements are difficult to distort into complex visual forms.
A canvas gives designers more freedom to draw and transform what appears on the page. Its contents, however, are treated by the browser as a flat image. Text inside a conventional canvas cannot be selected, and links there cannot be clicked.
HTML-in-Canvas connects these approaches. It places actual web elements within canvas graphics, so a page can use unusual shapes and effects without turning its text and controls into a static picture. The result can preserve the ability to interact with and select the content.
Examples range from knitting to 3D layouts
A demo page presents 15 examples of the approach. They include websites rendered as knitting patterns, 3D cubes and exploded layer views, showing how ordinary page content can be arranged inside more elaborate graphics.
That distinction matters for how people use a page. A visually transformed design can still contain real text that visitors can highlight and copy. Buttons and other embedded elements can also remain interactive, rather than functioning only as details in an illustration.
Felix Rieseberg, who works on Claude at Anthropic, wrote on X that Claude can build "bizarre & beautiful sites" with the feature. He said it reminded him of the Flash era. The comparison points to a familiar ambition: making websites feel like rich visual experiences, while using HTML elements that remain part of the page.
Support depends on the browser and app version
HTML-in-Canvas is being developed by Google as a web standard intended to preserve interactivity and accessibility. Browser support is not universal yet. The feature works in Chrome and Edge starting with version 157, while Safari and Firefox do not support it yet.
Claude's desktop apps support the capability from version 5.11492. To use it, people need to ask Claude for HTML-in-Canvas explicitly in their prompt. That makes the instruction part of creating the page: users should name the feature when they want Claude to build a graphic that contains real web elements.
These requirements shape where the output can be used. A page made with the feature may work in one supported browser and not in another, so browser compatibility matters when sharing or viewing the result. The underlying idea is to retain the visual freedom of canvas while keeping text and controls usable as web content.
A broader push toward visual AI work
HTML-in-Canvas is one example of Claude taking on more creative visual tasks. The source also reports that Claude Fable 5.2 painted photorealistic artwork stroke by stroke using self-written Python code.
Together, these examples show two different routes to visual output: generating artwork through code, and building web pages whose elements sit inside complex graphics. For the latter, the key change is that a page can look more like an illustration without giving up the practical qualities of selectable text and interactive controls.