A campaign image rarely stays untouched. Headlines change, layouts move and another format arrives. Canva’s Craft image model approaches generation as part of that continuing design work. Developed by Canva Research, it is available through Dream Lab and a public developer API.
Composition is a separate step
Craft first plans the image, including the elements, their relative size and placement, visual hierarchy and text. A rendering model then turns that plan into the image. That makes the intended arrangement a useful part of a brief: where should attention land, and which area needs to remain clear?
The first output can become working material
Canva highlights the connection with Magic Layers, which can turn generated imagery into editable design elements. For a team preparing several versions of a campaign, that puts the next edit into the conversation early. The question shifts from choosing one attractive result to deciding which parts must remain adjustable.
Format and language deserve their own decisions
Craft supports output up to 2K and more than 30 aspect ratios. Prompts can be written in different languages; text generated inside images supports 55 languages. The language used to describe a design and the language printed within it are separate choices.
One possible workflow is to brief a square social asset and a wide banner separately, keeping the central idea but changing its placement for each frame.
Hazel’s assessment
Where does the eye go first? Answering that before generation can make the later editing work more coherent. Keep the visual idea clear, identify the wording that will change, and leave room for those changes. Craft’s planning-and-editing direction is interesting because it treats the first image as material to develop, rather than the entire finished design.
