The AI image tools most people know are good at pictures and bad at posters: the text comes out garbled and the layout falls apart. Canva is betting it can fix that. On 22 October it released Canva Craft, its first image generation model built in-house, TechCrunch reports.
The difference is what it learned from. Instead of training mostly on photographs, Canva trained Craft on the human-designed templates in its own library: posters, social posts, presentations, infographics, menus and typography-heavy layouts.
How it works
- Craft first reads the prompt and makes a plan: where elements sit, how big they are, what matters most and what text to write.
- It then renders an image of up to 2K resolution, in any of 30 aspect ratios for different platforms.
- Canva says this gives accurate details and legible text, and lets you edit one part without changing the rest.
- Images can be turned into layered, editable designs with Canva Magic, the company's image-to-design model.
Craft is live for all Canva users in the Dream Lab tab and through the Canva API. Canva plans to bring it to ChatGPT and Gemini through its MCP connector, and says two cheaper, faster versions, Craftfast and Craft-ultrafast, are coming soon. Researchers from Leonardo.ai, which Canva bought in 2024, worked on the model.
It also helps Canva's costs. The company has relied on several outside image models, and its own should cut what it pays for them, TechCrunch notes. Canva has cut its growth forecast for 2026, and outside analysts say its valuation has slipped from $42 billion to $35 billion.
