Canva has released its first image generation model, Canva Craft, developed by its research team and contributors.
The model uses a two-stage prompt-to-plan pipeline to generate images up to 2K resolution across 30 aspect ratios, trained on human-designed library templates.
According to the company, its operation is based on two stages. First, the system reads the user's instructions and draws up a drawing with details about the position of objects, scale, visual hierarchy and text creation. Then, it produces the image in up to 2K resolution, supporting about 30 different dimensions.
How the model was trained
To train Canva Craft, the company didn’t use simple photos, but rather pre-made designs and templates from its library, such as posters, presentations, menus, and social media posts. This approach allows the system to better render details and clean text in images, which can then be transformed into layered designs within the platform.
Availability and next steps
Canva Craft is available to all users through the platform's Dream Lab tab, as well as through a dedicated program communication link. The company also plans to enable image creation in third-party chat apps, such as ChatGPT and Gemini.
READ ALSO
End of an era for the Arc browser, which will close permanently on December 22
Researchers from Leonardo.ai, which Canva acquired in 2024, also contributed to the development of the model. At the same time, the company announced that it is preparing two even faster and more economical versions, called Craftfast and Craft-ultrafast, for future release.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)