Use the model’s layout format
A layout includes summary.caption and an items array. Each item has a label, caption, and bbox; text items also include text. Set bbox_format to yxyx and bbox_coord_space to abs_centralized. The generator’s JSON mode includes an editable starter.
Think from the center
A bounding box is [top, left, bottom, right] in pixels relative to the canvas center. On a 1024-pixel square canvas, each axis spans −512 to +512. Keep boxes inside the selected canvas. The interface validates the structure before submitting.
Treat layout as direction
Start with two or three elements. Leave space between text blocks and avoid tiny lettering. The output is still a generated image, so inspect spelling and placement. Change one bounding box at a time when refining a design.
Go to the source
Official references, checked October 21, 2026.
Official layout specification