Canva Launches Neural Network for Image Generation
The product is named Canva Craft
The graphic design platform Canva has announced the release of its first proprietary neural network for image generation, called Canva Craft. The model was developed by the company's internal research team with the participation of specialists from the previously acquired startup Leonardo.ai and is intended to directly compete with counterparts from Google, OpenAI, Meta, and ByteDance.
The main difference between Canva Craft and classic image generators is its focus on professional layout and design. The model operates in two stages: - Planning: The neural network first analyzes the text prompt and creates a schematic of the future composition, calculating visual hierarchy, scale, precise spatial relationships between elements, and text layout. - Generation: Based on the plan, the final image is created with a resolution of up to 2K in 30 different aspect ratios for any platform.
The key feature stems from the dataset: Instead of a traditional photo database, Canva Craft was trained on real templates from the service's library—posters, presentations, infographics, and menus. As a result, generated images can be instantly broken down into separate editable layers using the Canva Magic tool.
The Canva Craft model is already available to all platform users in the Dream Lab tab and will also be accessible via the Canva API. It will be supported in third-party chatbots (including ChatGPT and Gemini) through the MCP protocol. Soon, the service will be enhanced with two accelerated modifications—Craft-fast and Craft-ultrafast.
The development of its own AI comes amid a business reassessment: Last year, the platform's annual revenue approached $4 billion. However, a reduction in growth forecasts for 2026 led independent analysts to lower Canva's valuation from $42 billion to $35 billion.
