Canva has launched Canva Craft, its first proprietary AI image model, globally for Canva users. Its two-stage design is intended to plan a composition before generating it, bringing the model closer to the way designers arrange a page than to a prompt-in, picture-out slot machine.
Watch Desk analysis
What happened
Canva Craft is built specifically for design work. CNET reports that the model can plan an image’s layout before generating it, and that users can edit the results. Canva’s head of AI research, Stefano Corazza, told the publication that the model was trained on Canva’s library, licensed and public datasets, and material from users who opted in to training.
CNET also cites early DesignArena results placing Canva Craft fourth among image models, behind OpenAI’s ChatGPT Images and ahead of Google’s Nano Banana and Microsoft’s MAI-Image. Those are early benchmark results, not a guarantee that it will excel at every kind of design.
Why it matters
The pitch is not simply that Canva has another image generator. It is trying to make the model build a composition that can then be adjusted, an important distinction for people who need a usable design rather than an attractive image they cannot sensibly edit. Canva says keeping the technology in-house should also reduce cloud costs and inference time.
The launch gives Canva a more direct role in building the AI tools inside its design platform, rather than relying only on third-party models. That raises familiar questions about training data and the place of working artists, alongside the practical question of whether the results are genuinely easier to shape.
Our read
Planning first and editing afterwards is a promising direction for design AI: fewer finished-looking images that fall apart the moment someone asks for a different layout. Canva Craft’s benchmark position is encouraging, but early rankings are a starting point, not a substitute for trying it on real briefs. The useful test is whether designers can get to a controllable, finished result faster, without surrendering the craft that makes the work good.
What to watch
- How Canva Craft performs on real design tasks, beyond early benchmark rankings.
- Whether its editing tools make generated layouts meaningfully easier to refine.
- How Canva explains the data used to train the model and the limits of user opt-in.
Discussion spark: Would a model that plans and lets you edit a design make AI genuinely useful to designers, or does it still leave the hard creative work to people?
Sources and evidence
- Source update (26 October 2026, 14:33 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.