Canva launches Craft, its first proprietary AI image model

Canva has released Canva Craft, its first in-house AI image generation model developed by Canva Research. Unlike many AI generators that focus on photographic realism, Craft was trained on millions of human-designed templates, including posters, presentations, and social media graphics, to better understand design principles like visual hierarchy and layout.

The model utilizes a unique two-stage architecture. An internal planning model first interprets a user's prompt to map out spatial relationships, element placement, and text requirements. A second rendering engine then converts this structural blueprint into a finished image with up to 2K resolution and more than 30 aspect ratios.

This proprietary approach aims to provide designers with greater creative control, more legible typography, and lower operational costs compared to using third-party AI tools. The model is available globally through Canva’s Dream Lab and to developers via a public API. Early benchmarks on DesignArena placed the model fourth overall, ranking highly in categories such as portraits and landscapes.