Canva has introduced Canva Craft, its proprietary AI image generation model, entering direct competition with offerings from OpenAI, Google, and Microsoft. The model employs a two-stage architecture that plans visual designs and text placement before final image generation, enabling increased precision, structural control, and layout consistency across up to 2K resolution and 30 aspect ratios.
According to Canva, the model was trained on millions of human-made design templates from its library, including social media posts, presentations, infographics, and menus. Canva Craft currently holds fourth place on the Design Arena crowdsourced benchmark leaderboard, trailing only variants of OpenAI’s GPT-Image-2.5.
To expand its ecosystem, Canva announced plans to support AI agent integration with Canva Craft via the open Model Context Protocol. The company also intends to release faster iterations, designated as Craft-fast and Craft-ultrafast, to lower latency for users worldwide through its platform.
Why it matters
Canva is verticalizing its stack with proprietary models trained on platform template data, reducing reliance on third-party API providers.
Integrating the Model Context Protocol allows third-party AI agents to directly program and manipulate visual design workflows.
Two-stage layout-first image models represent a key architectural trend for generating reliable text and structured design assets.
Source: theverge.com


