Canva releases Craft, its first in-house image model, which plans a composition before rendering it

On October 22, Canva introduced Craft, its first image generation model, built and trained from scratch by Canva Research, its in-house AI lab. According to BetaNews, the lab has more than 160 researchers and engineers and was formed in 2025 with teams from Kaleido and Leonardo.ai, two companies Canva acquired. Craft works in two stages. A planning model first interprets the prompt and writes a plan covering elements, spatial relationships, relative scale, visual hierarchy and any text. A separate rendering model then turns that plan into an image at up to 2K across more than 30 aspect ratios.

Canva says that, unlike rivals trained mostly on photographs, Craft’s training data includes millions of human-made templates from its library. The model was then refined with supervised fine-tuning and reinforcement learning using feedback from Canva’s designers. Per Canva, Craft debuted fourth overall on DesignArena and ranked first in People & Portraits and Landscapes & Nature. BetaNews reported that its score fell from 1340 to 1330 by October 23, with three OpenAI models ranked higher.

Craft is available to all Canva users globally through Dream Lab. Developers can use it in the Connect API, which is in preview, by setting the model name craft_v1_2k. According to Canva’s developer forum, each request uses the user’s AI credits and a job takes roughly 50 to 60 seconds. Support in the Canva MCP Server is coming soon. Cheaper Craft-fast and Craft-ultrafast variants are also planned, with no prices or dates announced.