Canva Launches Its First AI Image Model for Designers
Canva has launched Canva Craft, its first proprietary AI image model, available worldwide to all users. The tool promises more complex creations, finer editing work, and reduced cloud costs for the company.
Canva has launched Canva Craft, its first proprietary artificial intelligence image model, available worldwide to all its users. The tool is designed for designers and relies on a two-step architecture, intended to reason before generating an image. The goal is twofold: to offer more refined visuals and to keep more of the technical stack in-house, also a way to contain cloud costs and reduce inference time.
A model designed to fit designers’ work
According to Stefano Corazza, head of AI research at Canva, the model was trained on Canva’s library, licensed and public databases, as well as user data from contributors who agreed to participate in training. The idea is to produce more complex compositions, with logic close to that of a human designer: organizing elements, building a visual hierarchy, then adjusting the whole. Canva summarizes this process as a way to "reason" before generating. The term is well chosen, even if the market would probably have preferred a less anglicized expression.
The company did not start from scratch. Its first creative tool, Magic Media, dates back to 2023, and Canva has strengthened its AI capabilities with the acquisition of Leonardo.Ai. This growth explains the choice of an in-house model rather than simply integrating additional third-party services. From a product perspective, this mainly allows Canva to better control the experience, timing, and technical dependencies.
Canva aims to fix the real flaw of image generators
The key point is not just to create an image, but to be able to edit it without starting over. Canva relies on tools like Magic Layers, which break down an AI creation into editable elements. Canva Craft extends this approach and allows adjusting results, whereas many generators provide a good first version and then stop. For marketing teams as well as designers, it is often at this stage that time is saved or lost.
Canva also claims to have obtained encouraging results in early tests. On DesignArena, its solution would rank fourth among image models, behind OpenAI’s ChatGPT Images and ahead of Google’s Nano Banana as well as Microsoft’s MAI-Image. The service also stands out in portraits and product visuals but remains behind in the marketing subcategory, a field heavily frequented by Canva users.
An offensive in a still contested market
This launch comes amid a tense context for generative creation tools. Photographers, illustrators, and artists have long contested the datasets used to train some models, the risks of image counterfeiting, and the pressure they exert on creative professions. Canva, like Adobe, defends the idea of an assistance tool rather than a substitute. Platforms move forward, but creators still have to prove their images are truly human-made. Strange times: sometimes you have to authenticate your own work.
In practice, Canva mainly seeks to lock in its professional use. By keeping the technology in-house, the company improves its potential margin, better controls performance, and can evolve editing functions at the same pace as its interface. For users, this means less friction between generation, modification, and publication. For the market, it confirms that design suites no longer just want to integrate AI: they want to build it themselves.
By the numbers
- 2023 — launch of Magic Media by Canva.
- 4e place — Canva Craft’s ranking on DesignArena, according to CNET.
- 2 étapes — model architecture, according to Canva.
- 1 acquisition — Leonardo.Ai, cited as a lever for AI growth.