Canva has launched its first internally developed AI image generation model, called Canva Craft, putting the Australian design giant in direct competition with similar technologies from Google, OpenAI, ByteDance and Meta.
The model is built for design work rather than simple photo generation. Canva says Craft can produce images with readable text, accurate object details and support for precise editing, and it allows users to convert AI-generated images into layered designs that can be edited inside Canva like any other template.
Craft generates images through a two-stage process. In the first stage, the model analyses the user’s prompt and creates a design plan covering object placement, relative sizes, visual hierarchy and text positioning. In the second stage, it renders the image from that plan, supporting resolutions of up to 2K across 30 different aspect ratios. According to the company, this approach improves accuracy and lets users edit individual elements without disturbing the rest of the composition.
What sets Canva Craft apart is its training data. Instead of relying mainly on photographs, Canva trained the model on human-created designs from its own template library, including posters, social media graphics, presentations, infographics, menus and typography-heavy layouts. The company says this teaches the model how text, images and other elements should be arranged within a finished design.
The new model is available to all users through Canva’s Dream Lab section, with support planned for external applications. It integrates with Canva Magic, the company’s image-to-design technology that turns generated images into editable layers, building on the AI design model Canva introduced in 2025 to understand individual design elements and layers.
Canva also plans to make Craft available through third-party AI assistants, including ChatGPT and Google’s Gemini, using its Model Context Protocol integration. That would let users generate images through supported chatbots while drawing on Canva’s design capabilities, and a separate API will let developers embed the technology into their own applications.
Two additional versions of the model are in the works, called Craftfast and Craft-ultrafast, both aimed at reducing generation costs and improving processing speed. The company has not announced release dates for either.
The move is a strategic one. Canva previously relied on multiple external AI image generation models, and building its own technology could cut API expenses while giving it greater control over image quality and performance. Researchers from Leonardo.ai, the AI media generation platform Canva acquired in 2024, contributed to Craft’s development. The launch comes as Canva faces rising AI infrastructure costs: the company was on track to generate roughly $4 billion in revenue in 2025, but later trimmed its 2026 growth forecast, with external analysts reporting its valuation slipped from $42 billion to $35 billion.
