# Compositor annotation layers
Compositor nodes work like simple Photoshop layers. They combine a background image with a text layer that can be positioned anywhere. Here they are used to label every reference image so that the prompt can refer to them as A, B and C.
image A.image B.image C.The labels written onto the pictures match the role names in master-prompt-with-reference-image-roles. The model can tie each prompt instruction to the right reference without relying on input order alone.
The labelled node layout is part of the fixed scaffold and is not touched between runs. In multi-mannequin-composition, several mannequins are merged in a compositor into a single image A before this labelling step.
The text labels placed in the corner of each reference are what let the master prompt say "image A controls pose and camera, image B carries the product detail". Without the labels, the prompt can only point at the references by position or vague description. See master-prompt-with-reference-image-roles for the wording of the roles.
The compositor also merges everything into one layout image. That is why the downstream workflow stays the same whatever the scene contains (see multi-mannequin-composition).