Canva has entered the AI image race with its own model. Canva Craft, released this week, is the design platform’s first image generation model, built by its in-house research team to compete with models from Google, OpenAI, ByteDance and Meta.
How Canva Craft works
Craft generates images in two stages:
- Planning: it interprets your prompt and sketches a plan covering how elements relate in space, their relative size, visual hierarchy and the text to be generated.
- Rendering: it creates the final image at up to 2K resolution, in as many as 30 aspect ratios to suit different platforms.
Trained on designs, not photos
The biggest difference is the training data. While most image models learn mainly from photographs, Canva says Craft was trained on human-made designs from its template library, including posters, social posts, presentations, infographics, menus and text-heavy layouts. That focus is meant to help it handle structure, clear text and polished compositions, which are common weak spots for general-purpose image generators. Canva describes it as its first foundation model built from scratch by Canva Research, with contributions from researchers at Leonardo.ai, which Canva acquired in 2024.
Where you can use it
- In Canva: Craft is available to all users through the Dream Lab tab.
- Layered designs: images can be converted into editable, layered designs with Canva Magic, so you can move elements and change text afterward.
- API access: developers can use Craft through Canva’s API.
- Chatbots: Canva plans to bring image generation to third-party assistants such as ChatGPT and Gemini through its MCP integration.
- Coming soon: targeted editing to change one part of an image while leaving the rest intact, plus two cheaper, faster versions called Craftfast and Craft-ultrafast.
How does it rank?
According to a report on the launch, Craft debuted fourth overall on the DesignArena leaderboard, behind three OpenAI models. The same report says it placed first among 87 models for people and portraits, first among 85 for landscapes and nature, and second for abstract designs and patterns. These are early, third-party-reported results, so treat them as a first impression.
Why it matters
Canva has relied on several third-party image models, so building its own should give it more control over quality, speed and costs, although the company has not said how much it expects to save. For users, the pitch is images that start life as design-ready layouts rather than flat pictures you have to rebuild by hand.
Leave a Reply