The model generates images at up to 2KResolution across more than 30 aspect ratios, with a focus on design accuracy, text rendering and editable outputs.

According to Canva, Craft debuted as the fourth most advanced image generation model on DesignArena, a crowdsourced platform for evaluating AI design tools. The company also claims that Canva Research is now the world's second most advanced AI lab for image generation, based on the platform's rankings.

Craft ranks first in the People and Portraits category, which includes 87 models, and Landscapes and Nature, which covers 85 models. It also ranks second for Abstract and Patterns and fourth for Products, according to Canva.

Unlike conventional image generation models that generate pixels directly from a prompt, Craft uses a two-stage architecture that separates composition planning from image rendering. The first stage interprets a prompt and develops a plan covering elements in the image, their placement, relative sizes, visual hierarchy and any text that needs to appear. A dedicated rendering model then converts that plan into a finished image at up to 2K resolution.

Canva says this approach helps the model produce more accurate layouts, legible text and consistent object placement. It also allows users to make targeted changes to individual elements without unnecessarily altering the rest of the image. The architecture is designed to work with Canva's Magic Layers feature, which can turn generated images into editable design elements.

The company has trained Craft with a stronger emphasis on design than models primarily focused on photography. Its training data includes millions of human-made templates from Canva's content library, covering posters, social media posts, presentations, infographics, menus and typography-heavy layouts. Canva says these templates help the model learn how layout, visual hierarchy and typography work together.

After pre-training, the model underwent supervised fine-tuning and reinforcement learning, incorporating feedback from Canva's professional designers. For internal evaluations, designers and researchers assessed more than 25,000 pairs of images against competing models. Canva says it prioritised designers' judgments when they differed from automated evaluation results.

Craft further also supports a range of creative tasks, including marketing materials, social media graphics, editorial photography, illustrations, digital art, portraits, landscapes and conceptual imagery. Users can select from more than 30 aspect ratios or enter custom dimensions. The model also supports prompts in any language and can generate text in 55 languages, according to Canva.

The company has also focused on improving generation speed and efficiency. Canva says it used model distillation to reduce the number of inference steps required and developed a custom inference engine to optimise the planning stage. Two faster variants, Craft-fast and Craft-ultrafast, are in development and are intended to deliver image generation at a lower cost.

Canva Craft is available to users worldwide through Dream Lab, accessible from the bottom-left panel in Canva. Developers and product teams can also access the model through Canva's public API to integrate image generation into marketing platforms, e-commerce services, advertising tools and print services. Support through the Canva MCP Server is coming soon, allowing AI assistants and automated workflows to generate images for use in designs.

As with Canva's other AI products, Craft is covered by Canva Shield, the company's AI safety framework. It includes input and output moderation, privacy and safety controls, and reporting tools for potentially harmful or misleading content. Canva says it also evaluates models for brand safety, cultural context and harmful content before release.

Canva says future updates will focus on improving image quality, detail and editing control, including better handling of fine print and technical drawings.