Canva releases Craft model

Canva introduced Craft, its first in-house image generation model from Canva Research, which first plans the composition and then renders images up to 2K with readable text, securing 4th place on the DesignArena leaderboard.

Compiled by Sergey KostenchukPublished 2026-10-27Updated 2026-10-27

2026-10-27 Multimodal OpenAI

🎨 Canva releases Craft model

The Canva Research model operates in two stages: it first plans the composition—arranging objects and text blocks—and then renders the frame up to 2K in over 30 aspect ratios. Training on millions of Canva templates taught it to write readable text, and it has already reached fourth place on the DesignArena leaderboard.

🌍 Canva is entering generative AI with a focus on design: the "plan, then render" approach addresses typical competitor issues such as distorted text and broken compositions. Through MCP integration with ChatGPT and Gemini, AI agents will be able to invoke the generator.

👤 Craft is already available in Dream Lab and via a public API, while Magic Layers breaks down the result into editable layers. Fast Craft-fast and Craft-ultrafast versions are coming soon.

Source 1: https://www.theverge.com/ai-artificial-intelligence/1017151/canva-craft-in-house-ai-model Source 2: https://www.digitaltrends.com/computing/canva-craft-ai-image-generation-model-launch/

Sources