Canva has officially entered the competitive generative AI image market with the rollout of Craft, its first proprietary image generation model. Debuting at number four on the DesignArena leaderboard, Craft outperformed several prominent models from major tech companies while placing behind only three leading models from OpenAI. This strong market entry positions Canva Research as a high-performing lab in the image generation space following a year of substantial internal growth and investment.
Unlike standard text-to-image models trained primarily on flat photographic inputs, Craft was built using millions of professional human-made templates from Canva's extensive library. This unique training methodology equips the system with an intrinsic understanding of layout, visual hierarchy, typography, and structural composition. The model plans visual elements before rendering output, ensuring that graphics, text placements, and marketing materials maintain a cohesive and readable design standard.
The strategic shift toward in-house model development provides Canva with greater cost control and independence from third-party generative AI providers. Beyond powering internal design features, the company is making Craft accessible to external developers via an API integration. As the internal research division continues development on future iterations, upcoming updates aim to deliver faster processing times and more cost-effective options for enterprise and everyday users alike.
Comments