Canva has launched its first self-developed AI image generation model, Canva Craft, as the design platform pushes deeper into a market already crowded with offerings from OpenAI, Microsoft and Google.
Craft uses a two-stage architecture that plans a design before rendering the final image, which Canva says allows for "greater precision, consistency and control" — particularly for images that need legible text or carefully placed objects, a persistent weak point for AI image generators. The model can produce images at up to 2K resolution across more than 30 aspect ratios, and is aimed at everything from social media posts and marketing material to digital art and realistic scenes.
Trained on Canva's own template library
Canva says it trained Craft on millions of human-made templates drawn from its own content library, including social posts, presentations, infographics and menus. The model currently ranks fourth on Design Arena's crowdsourced leaderboard for AI image generators, behind three versions of OpenAI's GPT-Image-2.5 line.
The launch builds on a broader AI push at Canva, which introduced prompt-based editing and layered AI-generated images earlier this year. The company says Craft includes input and output moderation, privacy controls and reporting tools, and plans to let AI agents plug into the model through the open Model Context Protocol. Faster "Craft-fast" and "Craft-ultrafast" versions are planned for release later. Craft is available now to all Canva users worldwide. The push comes as design platforms race to keep pace with foundation-model labs that have far larger compute budgets; Adobe, Canva's closest rival in professional design, has pursued a similar in-house strategy with its Firefly models rather than relying solely on outside AI partners.