Ask for a red product on a blue surface and you will sometimes get a purple product on a purple surface. The model applies the whole prompt to the whole image, so attributes migrate.
Regional prompting scopes them. This region is the product, that region is the background, and the words for each stay where they were put. For any composition with two strongly described elements it is the difference between the shot you asked for and an average of it.
It costs setup time, so it earns its place on repeated compositions rather than one-offs. On a nine-variant campaign with the same staging, it is set up once.
Why do colours from one part of my prompt appear elsewhere in the image?
Because the whole prompt is applied to the whole frame, so attributes migrate between elements. Regional prompting scopes each description to an area and stops the bleed.
Is it worth the setup for a single image?
Usually not. It pays off on repeated compositions, where the regions are defined once and reused across every variant.
