Canva launches Craft as its first in-house image model

Canva has released Canva Craft, its first foundation model developed entirely by its internal research lab. The image tool is now available globally across Canva's Dream Lab interface and through its public developer API.

Unlike traditional image generators that convert text prompts directly into pixels, Craft uses a two-stage method. In the first stage, a dedicated planning model maps out the composition. It determines the position of individual objects, establishes visual hierarchy, sorts relative sizing, and plans text placement. In the second stage, an image rendering engine turns that plan into an image at resolutions up to 2K across more than 30 aspect ratios.

Canva said this separation between planning and rendering addresses a long-running problem in automated design: illegible typography and awkward element positioning. The model supports prompts in multiple languages and can produce readable rendered text in 55 languages.

Editable designs rather than flat pictures

Canva trained the model on millions of design templates, such as posters, menus, infographics, slides, and social media graphics. The output is aimed at marketing and commercial layouts rather than artistic or photographic work.

Because the underlying system understands which objects belong where, it works directly with Canva's Magic Layers tool. Instead of giving users a flattened image file that cannot be modified easily, the system can break the generated result into distinct layers. Staff can select and adjust text headings, move product shots, or swap background colours without having to regenerate the entire image from scratch.

Canva reported that Craft ranked fourth overall on DesignArena, an independent comparison platform for AI design tools, while taking first place in the portraits and horizontal scenery categories.

What this means for Australian businesses

For local businesses that produce marketing materials in-house, the release offers several practical benefits:

• Staff can create campaign graphics with readable headlines, which cuts down time spent manually matching generated art to overlaid text boxes.
• Marketing teams can revise specific parts of an image, like price tags or callouts, while keeping the rest of an approved design intact.
• Small teams with limited design budgets can produce branded collateral across multiple formats, from square social posts to wide banners, without starting each asset from zero.
• Developers and internal IT teams can plug Craft into existing software workflows using Canva's Connect API to automate routine graphics production.