Canva just built its own image model, and it plans the layout before rendering the picture. Called Craft, it first works out where elements belong, their relative size, visual hierarchy and text. A separate model then renders the image. Canva trained it from scratch using data that includes millions of human-made templates, then refined it with feedback from professional designers. Craft generates images at up to 2K resolution across more than 30 aspect ratios. Its outputs can also pass through Magic Layers, which turns generated images into editable design elements. That means you can keep working on individual parts instead of regenerating the whole image. Available now globally in Canva's Dream Lab and through a public API.