Canva launches Craft, its first in-house image generation model

Canva has unveiled Canva Craft, its first image-generation foundation model developed and trained from scratch by its in-house AI research unit, as the design platform seeks to strengthen its artificial intelligence capabilities.

The company said Craft debuted as the fourth most advanced image-generation model on DesignArena, a crowdsourced platform for evaluating AI design models. It claimed the launch made Canva Research the world’s second most advanced AI lab for image generation, based on the platform’s rankings.

Available globally through Canva, Craft generates images at resolutions of up to 2K across more than 30 aspect ratios. It is designed to produce images with accurate text rendering, precise object placement and editable elements, targeting applications ranging from marketing materials and presentations to illustrations and editorial imagery.

Unlike conventional image-generation models that generate pixels directly from prompts, Craft uses a two-stage architecture that separates image composition planning from rendering, Canva said.

In the first stage, a planning model interprets the prompt and maps out the composition, including the placement and relative scale of objects, visual hierarchy and text. A second model converts this plan into the final image. The approach is intended to improve consistency and allow users to modify individual elements without changing the entire image.

Craft ranked first globally in the People and Portraits category, among 87 models, and in Landscapes and Nature, among 85 models, according to Canva. It ranked second in Abstract and Patterns and fourth in Products.

The model was trained using millions of human-made templates from Canva’s content library, including posters, social media posts, presentations, infographics, menus and typography-heavy layouts. Canva said it also used supervised fine-tuning and reinforcement learning, incorporating feedback from professional designers.

For its internal evaluations, the company’s designers worked with researchers to assess more than 25,000 pairs of images in head-to-head comparisons with leading models.

“We’re incredibly proud of what we’ve achieved with Craft, and how it blends design understanding with best in class image generation,” said JJ Fiasson, Head of Generative AI at Canva.

“Craft delivers exceptional results for jobs where the layouts and text need to be perfect, and renders people and landscapes with stunning quality,” he said, adding that the model was “a step forward for Canva’s AI experience”.

Craft is available globally through Canva’s Dream Lab and through a public API that allows developers to integrate the model into their own tools. Canva said it plans to introduce faster versions of the model and improve its handling of fine print and technical drawings.