Canva’s first AI image model plans designs before generating them

Canva has introduced its first proprietary AI image generation model, named Canva Craft. A key feature of this model is its two-stage architecture, which first plans a design's composition and layout before generating the final image. This approach is intended to provide users with more precise control and consistency compared to other text-to-image models.

A Two-Stage Process for Design Control

Canva Craft's architecture separates the design process into two distinct phases. First, it creates a plan or a blueprint for the image based on the user's prompt. This initial stage determines the layout and arrangement of elements. Only after this plan is established does the second stage generate the actual pixels of the image. This method aims to provide greater precision, consistency, and control, positioning it against image generation models from OpenAI, Microsoft, and Google.

Technical Foundations and Market Position

By developing its own model, Canva aims to offer a more integrated and controlled creative experience within its platform. The model is based on foundational concepts similar to SPADE for automating environment generation, which suggests a focus on structured, layout-aware image creation. This move allows Canva Craft to compete directly in the growing field of AI-powered creative tools. (Source: The Verge)

Additional sources: - importai.substack.com