Render a preview (0.1 credit per call)

Read one box of the page instead of the whole sheet, in CANVAS pixels (the coordinates your layers already use). THIS is the endpoint for it: the page is composed at its own resolution, the box is cut out of it and only then capped at 640 px on its long side, so a box no larger than that comes back 1:1 and small type on a 2480 px print sheet is finally legible. The reply carries region and scale (multiply canvas pixels by scale to land on the pixels you are looking at). Same price as any preview. Cannot be combined with debug_overlay, max_width or canvas.finish_profile; a box outside the canvas, or with zero area, is refused with the canvas size in the message.

Geometry of the successful render attempt, bound to job_id and attempt_id. Layout is the DSL coordinate space. Scale layout to pre_finish, map finish.source_rect to finish.destination_rect when present, then scale output to the actual displayed image. Without finish, pre_finish equals output. Display resizing never changes this report. Absent on legacy jobs or document results without raster geometry.

What this call took and what is left.

What the preview MEASURED for every text layer, when the render reports measurements

With region — the box that was cut out, in canvas pixels