Canva launches Craft image-generation AI model in Jakarta
The internally developed model is available through Dream Lab and a public API, with faster variants planned.
Canva introduced Canva Craft, an AI foundation model for generating images, in Jakarta on 30 October. Developed by its internal research laboratory, the model targets visual communication design and is accessible through Canva’s Dream Lab and a public API for developers.
According to Canva, Craft produces 2K images in more than 30 aspect ratios. Its two-stage architecture handles layout planning separately from final image rendering. The model was trained using millions of assets from Canva’s library to develop its handling of composition, typography and visual hierarchy.
JJ Fiasson, Canva’s Head of Generative AI, said Craft combines design understanding with image generation, including work that requires precise layouts and text, as well as images of people and landscapes.
Canva said images generated by Craft can be converted into editable layers. Structural information in JSON format allows its Magic Layers technology to identify individual elements, enabling users to change specific components without manually rebuilding the entire image.
The company said its researchers had used model distillation to reduce inference steps and speed up processing. Further optimisation is intended to produce variants called Craft-fast and Craft-ultrafast. Canva also said text and image prompts undergo moderation and continuing safety testing through Canva Shield.
Canva plans to offer developers access to Canva MCP, intended to connect Craft with external AI assistants for automated workflows.
