Canva has launched Canva Craft, an AI image model that plans a design before it generates anything, and the company is positioning it against image models from OpenAI, Microsoft, and Google.
The model uses a two-stage architecture: one step lays out the design, the next produces the image. In its announcement, Canva said this allows for "greater precision, consistency and control." That is the company's own description, and the source gives no outside testing of it.
The stated specs are easier to pin down. Canva Craft generates images at up to 2K resolution and in more than 30 aspect ratios. Canva said it is meant for social media posts and marketing content, and also for realistic scenes, digital art, and environmental imagery.
The plan-first approach is the interesting part. Canva sells design tools, not standalone pictures, and design work tends to hinge on composition: what goes where, and whether a set of assets looks like it belongs together. A model that settles the layout before rendering is a sensible fit for that job. The wide range of aspect ratios points the same way, since marketing material rarely comes in one shape.
It also marks a shift in Canva's position. Its first AI image model arrives in a field where three of the largest AI players already compete, so the pitch has to rest on something other than raw image quality. Resolution and aspect ratios can be checked by anyone. Whether planning first delivers more consistent results is a question for side-by-side use by Canva's customers.