Canva Releases Proprietary AI Image Model Called Canva Craft

Canva has launched its first proprietary AI image model, named Canva Craft, aimed specifically at designers and available globally to users. According to the company, the release follows years of integrating third-party AI tools and builds on internal research and the acquisition of Leonardo.Ai.

How Canva Craft Works

• Two-Stage Architecture: The model uses a two-stage approach combined with prompt-improvement tools to reason before generating visuals.
• Design Process Mimicry: Executives state that the model plans layouts and elements to replicate a professional designer's workflow.
• Training Data: Canva reports the model is trained on its own library, licensed and public databases, and data from users who opted in.

Performance and Editing Capabilities

• Benchmarking: Canva reports positive early results on DesignArena, placing fourth overall among image models.
• Post-Generation Editing: The model integrates with editing features like Magic Layers, allowing users to modify individual elements within AI-generated designs.

Industry Context and Concerns

Generative AI tools continue to face scrutiny from photographers and artists regarding copyright concerns and the broader impact on creative professions. Industry figures have publicly debated the role of AI versus human creators, while traditional creators navigate the challenges of digital verification and attribution.

How to Verify

Users can check the availability and updates regarding Canva Craft by visiting Canva's official blog or checking the platform's in-app tool announcements.