Canva has released Canva Craft, its first image generation model, developed by its research team. Canva says the model can produce readable text and accurate object details, while letting users make precise edits to selected areas without changing the rest of an image.
Craft plans the composition before generating the image
According to Canva, generation follows two stages. Craft first interprets the user’s prompt and creates a plan specifying the positions and relative sizes of elements. That plan also covers visual hierarchy and the information needed to generate text.
In the second stage, Canva says, the model generates an image at a resolution of up to 2K. It supports 30 different aspect ratios for different platforms.
Trained on designs from Canva’s library
Canva says it trained Craft on human-created designs from its library rather than photographs. The training material ranged from posters and social media posts to presentations and infographics, alongside menus and typography-heavy layouts.
Researchers from Leonardo.ai helped develop the model, according to Canva. The company acquired the media generation platform in 2024.
Converting generated images into layered designs
Images generated with Craft can be converted into layered graphic designs using Canva Magic, a separate image-to-design tool.
Available in Dream Lab, with API and chatbot support planned
Canva Craft is available to all users on the platform through the Dream Lab tab. Access through the Canva API is set to follow.
Canva also plans to support image generation through its MCP in third-party chatbots such as ChatGPT and Gemini.
The company is developing two additional variants, Craftfast and Craft-ultrafast, which it describes as cost-efficient and plans to release soon.
