Canva launches Craft AI image generator to rival Google and Microsoft

A new entry in the AI race

The Australian technology company Canva has officially released its first public AI model, named Canva Craft. This new tool is an image generator designed to create visual content directly within the company's ecosystem. It marks a significant step for the firm as it seeks to provide more integrated AI capabilities to its massive user base.

According to reports from NewsWire on MSN, independent testing by a site called Design Arena shows that Craft is performing exceptionally well. The rankings indicate that the tool is currently outranking major competitors including Google, Meta, Microsoft, xAI, and ByteDance. It specifically places above Microsoft’s MAI-Image and Google’s Nano Banana.

While it is a top performer, it is not the absolute leader in every category. The independent assessor noted that only three models from OpenAI currently rank higher than Craft. However, the tool still holds a strong position in the broader landscape of AI image generation.

Training on design, not photos

One of the most distinct features of Canva Craft is the specific data used to train the model. Instead of relying primarily on general photographs, the company trained the AI on millions of human-made design assets. These include social posts, infographics, presentations, and posters.

By focusing on these specific types of content, Canva aims to create an AI that understands design principles. The company wants the tool to be useful for people who need to create marketing materials and professional layouts. This specialized training helps the model align with the actual needs of Canva's existing customers.

The results of this training approach are visible in the independent rankings. Craft was placed at the No.1 spot for creating images of people, as well as for nature and landscapes. It also ranked second for abstracts and fourth for products, showing its versatility across different visual styles.

Solving the cost of AI

The decision to build a homegrown model was largely driven by financial necessity. Earlier this year, Canva was forced to delay the launch of its AI tools because of the high costs associated with letting non-paying users access third-party models. Specifically, using models from companies like OpenAI proved to be too expensive for the company's business model.

By developing Craft, Canva can manage its own infrastructure and expenses. A company spokesperson stated, "Building and owning these models gives Canva complete control over cost, quality, and speed." This independence allows the company to offer AI features to hundreds of millions of monthly users without the same financial burden.

The pressure to solve these cost issues was significant for the company's market position. During the period when the AI launch was postponed due to high costs, the company saw $10 billion wiped from its valuation. Launching Craft is a direct response to the need for a more sustainable and profitable AI strategy.

Growth of Canva Research

The release of Craft is the first public output from a dedicated team known as Canva Research. This internal group has seen rapid expansion over the last year, tripling its staff numbers to keep up with the demand for new technology. The team is currently working on a pipeline of additional models that will be released in the future.

This expansion suggests that Canva is moving toward a more robust internal AI laboratory. Rather than just being a platform that plugs into other people's tech, they are becoming a creator of the underlying technology. This shift allows them to customize the AI's behavior to fit their specific brand and user experience.

The company is also opening up these capabilities to a wider audience. Canva Craft will be made available to external developers and AI agents. This move could allow other software and automated systems to integrate Canva's specific design-focused AI into their own workflows.

What comes next for users

For the average user, the launch of Craft means more reliable and affordable image generation within the tools they already use. Because the model is optimized for design, users can expect outputs that look more like professional marketing assets and less like generic AI art. It bridges the gap between raw generation and finished design.

The company's strategy of owning the