Canva is making strides in the competition for artificial intelligence-generated imagery. On Thursday, the design platform introduced Canva Craft, its inaugural AI image generation model, aimed at competing with offerings from major players like Google, Meta, and OpenAI.
Craft can be accessed via Canva’s Dream Lab, even by users with free accounts, although it comes with certain usage restrictions concerning AI. Users must log in to their Canva accounts to begin using the tool.
In its launch announcement, Canva revealed that the model takes into account a design’s layout, proportions, and text before generating an image. This strategy is intended to produce text that is clearer and more detailed than offerings from other platforms.
The training data for this model consisted of templates originally created by humans on Canva, which include various designs such as posters, menus, and presentations, rather than relying heavily on photographic images.
Business Insider decided to assess Canva Craft's performance by creating three image generation prompts that progressively increased in complexity—from a straightforward greeting card to a highly realistic image generated entirely from scratch.
The first task involved creating a birthday card. The prompt asked for a cartoon-style "Happy Birthday" card designed for a friend named Jessica, who has distinctive pink hair and brown eyes, is fond of paisley patterns, and will turn 28 in November.
In this instance, Canva's rendering of the birthday card did not quite meet expectations, suffering from a mismatch in font choices and the cartoon character's oversized head, and it notably omitted the paisley detail altogether. In contrast, both ChatGPT and Muse managed to capture the essence of the request, with ChatGPT suggesting pink as a possible favorite color for Jessica and Muse incorporating warm, autumnal details into the design.
In the second test, the goal was to edit an original image featuring the author holding a necklace pendant. The task required the AI models to remove the red garnet pendant from the model's hand and position it as if it were displayed in a jewelry showcase. Canva successfully identified the pendant and generated a creative backdrop, but both it and Gemini created physically improbable scenarios, such as depicting the pendant as floating or awkwardly balanced on a pole. ChatGPT and Muse excelled in this task as well, with ChatGPT standing out for maintaining realism, despite Muse’s unrequested alterations to the pendant.
The third and final prompt asked for the creation of a photorealistic image of a long-haired orange tabby cat in a Halloween costume, complete with bat wings, sitting atop a gothic castle beneath a full moon and a starless sky. Although Canva’s depiction exhibited some exaggerated elements, like overly striking fur detail, it largely met the task's requirements and matched the quality of images produced by Muse and Gemini. However, ChatGPT emerged as the clear favorite due to its lifelike portrayal of the cat and well-executed lighting that enhanced the scene's realism.
In summary, while Canva is taking initial steps into AI-driven image generation, it still has significant improvement to make when compared to established models. For the time being, users may find it more effective to engage in hands-on manipulation for their design needs.
