Canva Launches Its Own AI Image Generator to Take on Google and OpenAI

Canva has recently unveiled its own AI image generation model, called Canva Craft, as part of its strategy to rival rivals such as Google, OpenAI, ByteDance, and Meta. This new technology is capable of creating images that include readable text, precise object details, and the ability to make accurate edits. Users can also convert these AI-generated images into layered designs that can be further edited within the Canva platform.

Canva Craft is accessible to all users through the Dream Lab section of the platform, with further integration planned for external applications. The functioning of Canva Craft involves a two-stage process. Initially, the model examines the user's text prompts to devise a design plan, taking into account factors like object placement, relative sizes, visual hierarchy, and text positioning.

The second stage then generates the image based on this plan, with support for resolutions up to 2K and 30 different aspect ratios. According to Canva, this method enhances image accuracy and allows users to edit individual elements without impacting the entire image. A key feature of Canva Craft is its use of human-designed templates for training, as opposed to relying heavily on photographs.

The model has been trained using design templates from posters, social media graphics, presentations, infographics, menus, and typography-heavy layouts, enabling it to comprehend how text, images, and other elements should be arranged within a design. This model can also be integrated with Canva Magic, Canva's existing image-to-design technology, allowing generated images to be converted into editable layers.

Furthermore, Canva plans to make Craft accessible through third-party AI assistants like ChatGPT and Google's Gemini, using its Model Context Protocol (MCP) integration. Additionally, Canva aims to offer Canva Craft via its API, enabling developers to incorporate the technology into their applications.