Canva unveils Craft, a model that plans the layout before generating an image

Canva has unveiled Craft, its own image-generation model for design layouts. According to an October 27, 2026, article on Habr, the model plans the composition and text placement before generating an image. The result can then be sent to Magic Layers, where individual elements can be edited without regenerating the entire image.

How Craft builds an image

Craft was developed from scratch by the Canva Research team. The model is designed not just for illustrations and photorealistic scenes, but also for materials where text and layout matter: posters, presentations and ad banners. For these tasks, a suitable image isn't enough. The headline needs room, and objects must not get in the way of the message.

The process has two stages. First, one model determines the positions of objects and text blocks, the scale of elements and the visual hierarchy. Then another generates an image based on that plan. Canva intends this approach to help avoid problems with object placement, proportions and typography. But the approach itself doesn't guarantee that every result will work without edits.

Craft was trained on millions of Canva user templates, then fine-tuned with input from professional designers. That helps explain why the model's description focuses on the structure of the finished design, not just the image's style.

Craft supports resolutions up to 2K and more than 30 aspect ratios, including custom ones. The model can generate text within images in 55 languages. Those specifications make it possible to create materials in different formats, but they don't show how well the model will place a particular headline or a substantial block of text.

Editing, access and Canva's plans

A generated image can be sent to Magic Layers to separate it into editable elements. If the headline or an object's position doesn't work, a designer can change it in the editor instead of regenerating the whole image. For a layout with several elements that already work, that's more useful than starting every revision with a new prompt.

Craft is already available to try in Dream Lab. Developers can access an API under the model ID craft_v1_2k; generations through it use the Canva user's AI credits. The article provides no pricing or limit information, so it can't be used to calculate the cost of producing a large number of layouts.

According to Canva, Craft ranked fourth overall on DesignArena at launch and first in the portrait and landscape categories. Those rankings don't tell us how the model will handle a specific working layout. The company also promises Craft-fast and Craft-ultrafast versions, along with integration with Canva MCP, which would let AI agents create images in automated workflows. No release dates have been given for these features.

What this means for websites and marketing

For a marketer or website owner, Craft could be a way to draft a banner or another piece of content with text. If space for the message is planned before the image is generated, it's easier to assess whether the composition fits the task. The finished version still needs to be checked for text accuracy, readability and whether the main point stands out clearly enough.

For designers, the more interesting part is the combination of an upfront plan and editable elements. You can start with the overall structure, then adjust the headline or an object's position while keeping the parts that work. That could help when comparing versions of the same design. Editing individual elements also brings to mind another approach to working on the canvas: Code Layers let you add code in Figma, though Craft is focused on image generation.

Developers have access to the API; generations through it use the Canva user's AI credits. Canva MCP integration has only been announced as a future feature, not as an available step in the workflow. A team assessing the time and credits required would be better off testing a small use case first than building a process around promised versions.

My take: check the hierarchy first

I like the idea of deciding where elements go before generating the image. When working on practice layouts, I check the mobile screen early: that's where it becomes clear faster if a headline is too long or a secondary block has taken up space meant for the main message. If I can edit parts of a Craft result separately, that's a useful starting point for me.

But I wouldn't treat a generated banner as a finished design. On a small screen, I find it easier to judge whether the text is readable and the order of elements makes sense. Clear navigation matters more than elaborate transitions in a portfolio; I follow a similar principle in graphics: make the message clear first, then work on the styling.

Sources

Where the news comes from. The text is a retelling in the author’s own words; the facts come from the source, the opinion is the author’s.

1. Canva выпустила Craft: нейросеть, которая сначала продумывает дизайн, а потом генерирует картинку habr.com

About the author

Ella Jäger

Aspiring web designer

I live in Berlin, study interface design, and build my portfolio with small assignments. Right now I’m checking my mockups on a phone before showing them to my mentor.

All posts by the author