Forum
Canva AI Design Tools: Who Drew the Templates Craft Learned From?

Canva’s AI design tools gained a new engine on 22 October, when Canva introduced Canva Craft, the first image generation model built by its own research team. Canva says Craft was trained on human-made templates from its content library rather than mainly on photographs. For a platform whose business rests on layouts made by designers, that detail is the most interesting part of the launch, and it raises a question the launch coverage leaves open: who designed the material the model learned from?
What Canva announced
According to TechCrunch’s report, Craft was trained on templates such as posters, social posts, presentations, infographics, menus and layouts with a lot of lettering. It plans the composition first and then renders the image, produces output up to 2K resolution in more than 30 aspect ratios, and is available in Canva’s Dream Lab and through its API. Canva also plans to offer it through its MCP server, and says two cheaper variants, Craftfast and Craft-ultrafast, are coming. The company reports that Craft debuted fourth on the DesignArena ranking, a figure that is Canva’s own and preliminary.
This is a different step from the one the platform took in April, when Canva AI 2.0 and the arrival of readable typography showed generative tools learning the grammar of layout. It is also separate from the cost story in August, when Canva slowed its AI rollout because serving AI features cost more than expected. The new point is ownership of the model, and of the data behind it.
Canva AI design tools and the people behind the templates
A template is not raw material in the way that a pile of photographs is. Someone chose the typeface, set the spacing, balanced the colours and decided where the eye should go first. When a model learns from thousands of such layouts, it learns those decisions. The launch materials describe the training data, but the open question is what terms cover the people who made those templates, and whether they were asked.
The same week, the issue surfaced from the other side. The illustrator Axel Scheffler, known for the pictures in The Gruffalo, told the German weekly Die Zeit that he is considering legal action against AI companies after chatbots produced images of his character that he found strikingly close to his drawings, according to reports from the German news service epd. He is considering legal action, nothing more at this stage. His complaint is still instructive. A creator’s work is recognisable at a glance, and the people who made it want a say in how it is reused.
Licensed, consented or merely available
Those two stories sit at opposite ends of a spectrum. A platform that trains on its own library is on firmer ground than a company that gathers images from the open web, because it has a relationship with its contributors and terms it can point to. Yet a relationship is not the same as a clear agreement to train a model. Earlier columns on Getty’s display-only licence with OpenAI and on why paying artists per output may never work showed how many layers of permission now sit between a work and a generated image.
What designers can do now
- Read the contributor terms of any platform where you publish templates, and note whether they mention AI training or compensation.
- Ask suppliers which model produced a layout, and what that model was trained on.
- Keep your own source files, so that you can show which parts of a design are your work.
- Disclose when a layout began as model output, in the same way you would credit a stock template.
- Treat a polished first result as a draft, since a model trained on popular templates tends to return the average of them.
The award perspective
The International AI Award Rules page asks entrants to explain how AI entered their process, which models and tools they used and what choices they made. For work built on a platform’s own model, that explanation should include the model and the role the entrant played, so that judges can tell design decisions from default output.
The takeaway
Canva Craft is a sensible product move, and training on a company’s own library is easier to defend than scraping. Still, the templates in that library carry the work of thousands of designers. As platforms build their own models, the question for creators is no longer only whether their work was used, but on what terms, and whether anyone asked.



