Canva has released its first in-house image generation model, a useful shift for a company that sells design workflow, not just AI tricks. The model, called Canva Craft, is available inside Dream Lab for all Canva users and will also be offered through the company’s API, according to TechCrunch.
The interesting part is less that Canva now has an image model — every large software platform seems contractually obliged to say that — and more what Canva says it trained it on. Rather than photographs, the company said Canva Craft was trained on designs from Canva’s own template library, including posters, presentations, menus, infographics and other layouts where text placement and visual hierarchy matter. That is a narrower target than the general-purpose image generators pushed by OpenAI, Google, Meta and ByteDance. It is also closer to where Canva actually makes money.
Canva said the model works in two stages: first interpreting a prompt into a structured plan covering elements such as spatial relationships, scale and text; then generating an image at up to 2K resolution across as many as 30 aspect ratios. The company says that setup improves object detail, makes text more legible and allows more precise edits without changing unrelated parts of an image.
Those claims are from Canva, and they are plausible rather than proven by the excerpt alone. But the product logic is clear enough. If a generated poster or social graphic can be converted into a layered design inside Canva’s editing tools, the company is not just offering another image bot. It is trying to keep users inside its own production line, from prompt to publishable asset.
That matters financially as well as technically. TechCrunch noted that Canva has already been using third-party image models on its platform, so an internal model could reduce API costs while giving Canva more control over speed and output quality. Cost control tends to look especially attractive when growth cools. TechCrunch reported that Canva had been on track for $4 billion in revenue last year, but that a cut to its 2026 growth forecast helped drive external analysts to lower its implied valuation from $42 billion to $35 billion.
Canva also said researchers from Leonardo.ai, the media generation startup it acquired in 2024, contributed to Craft. Two cheaper versions, Craftfast and Craft-ultrafast, are in development. Faster and cheaper is not poetry, but in software it often passes for strategy.
Dua Mifsud
Harvey Jr.
Ryan C