Canva launches own AI image generator to compete with Google and OpenAI

A shift toward internal tech

Canva is officially entering the custom AI model space with the launch of Canva Craft. This move positions the design platform as a direct competitor to major image generation technologies from companies like Google, OpenAI, ByteDance, and Meta. By developing its own model, Canva aims to gain more control over the specific performance and quality of the images it produces for its users.

Until now, the company has relied on various external AI image generation models to power its features. Moving to an internal solution allows Canva to manage its own technical roadmap more independently. This transition is a strategic response to the high costs associated with using third-party APIs for heavy AI tasks.

The new model is currently available to all users through the platform's Dream Lab section. While it is currently focused on the Canva ecosystem, the company has stated it plans to provide additional support for external applications in the future. This indicates a broader strategy to make Canva's proprietary technology a standard for design workflows.

Training on design, not photos

What sets Canva Craft apart from many other image generators is its unique training data. While many models rely heavily on general photographs, Canva trained Craft using human-created designs from its own template library. This includes a wide variety of assets such as social media graphics, presentations, infographics, and menus.

Because the model was trained on these specific types of content, it has a better understanding of how to arrange elements like typography and layouts. It is specifically designed to handle text-heavy designs, which is often a struggle for general-purpose image generators. This focus helps the model produce results that are more useful for professional and social media marketing.

The model also supports a wide range of technical specifications for its outputs. It can generate images in 30 different aspect ratios and at resolutions of up to 2K. These features are intended to ensure that the generated content fits the diverse needs of different design projects across various platforms.

How the two-stage process works

Canva Craft uses a specific two-stage process to turn a user's text prompt into a final image. In the first stage, the model analyzes the instructions to create a comprehensive design plan. This plan covers essential details like the placement of objects, their relative sizes, the visual hierarchy, and where text should be positioned.

Once the design plan is established, the second stage begins the actual generation of the image based on those specific instructions. This method is designed to improve image accuracy compared to models that try to generate everything at once. It ensures that the final output aligns more closely with the user's original intent and structural requirements.

One of the most useful features for designers is the ability to turn these images into layered designs. By integrating with Canva Magic, the company's image-to-design technology, users can convert a generated image into editable layers. This allows a user to modify individual elements of a design without affecting the rest of the image, a significant step up from static image generation.

Expanding reach and integration

Canva does not intend to keep Canva Craft restricted solely to its own website. The company plans to make the model available through third-party AI assistants, including Google’s Gemini and ChatGPT. This will be achieved using the Model Context Protocol (MCP) integration, which allows these chatbots to access Canva's specific design capabilities.

To support the broader developer community, Canva also plans to offer the model through an API. This will allow other developers to integrate Canva's image generation technology into their own applications. By opening up these pathways, Canva is positioning its model as a foundational tool for the wider AI and design industry.

The company is also looking toward future iterations to optimize performance. They are currently developing two new versions of the model called Craftfast and Craft-ultrafast. While no release dates have been set, these versions are specifically designed to improve processing speeds and further reduce the costs of generating images.

The business of AI costs

The move to build internal models is largely driven by the rising expenses of AI infrastructure. Canva was originally on track to generate approximately $4 billion in revenue in 2025, but it recently had to reduce its growth forecast for 2026. This adjustment was made due to concerns over the high costs associated with developing and maintaining AI technologies.

The financial impact of these costs has already been noted by market observers. Following the revised growth forecast, reports indicated that Canva's valuation declined from $42 billion to $35 billion. These figures highlight the pressure that high-end AI development puts on the company's bottom line and overall valuation.

By developing its own technology, Canva hopes to mitigate these risks by reducing its reliance on third-party providers. This strategy aims to give the company more control over its operational costs while maintaining high-quality performance. The involvement of researchers from Leonardo.ai, a platform Canva acquired in 2024, also played a key role in developing the Craft model.