Canva Craft: Canva’s first in-house image model, trained on templates
Canva released a model trained on human-made designs rather than photos. Results can become editable layers.
Canva has released Canva Craft, its first image generation model. The company previously relied on third-party models. Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the work.
What’s new
The model works in two stages: first it plans element placement, scale, hierarchy and text, then it generates images up to 2K in 30 aspect ratios.
The difference is the training data. Canva says it used human-designed templates from its library, such as posters, social posts, presentations, infographics and menus, rather than photographs. So the bet is on layouts and text. The company also claims precise edits that leave the rest of the image untouched, and Magic Layers can turn a result into a design with editable layers.
How to try it
The model is available to all Canva users in the Dream Lab tab and through the Canva API. Canva also plans image generation in third-party chatbots such as ChatGPT and Gemini through its MCP. Faster, cheaper variants of the model are due soon.
Limitations
Quality claims come from Canva itself. Test it on your own layouts, especially text in non-Latin scripts, which the coverage doesn’t address.
Prepared with AI from public sources.