Canva Releases Craft, Its First Image Generation Model

Canva has released Canva Craft, its first in-house image generation model. The platform's own research team built it to compete with models from Google, OpenAI, ByteDance, and Meta. The model works in two steps. First it builds a layout for the image, deciding where elements sit and how big they are. Then it renders the actual picture, at resolutions up to 2K and in thirty different aspect ratios. Canva did not train it on photographs. It trained the model on its own real templates instead, things like posters, social media posts, and presentations. That is why text in the images comes out accurate. It also means a single element can be edited without touching the rest.

There is another angle here for anyone building a product or content pipeline on top of several AI services. Canva itself frames this launch as a way to cut spending on third-party APIs it used to buy from other vendors for image generation. Accurate text and clean editing are exactly the things that usually break in image generators, and Canva says its separate planning stage fixes that. I would treat the claim with some skepticism, though. Other models have promised accurate text and clean editing before. What matters is how Craft actually performs on real layouts, not what the announcement says.