oh, I see what you mean. short answer - it literally repaints the image using the transformed data. For positional stuff (eg. eye size/position) it transforms UV space - in the same way as a flowmap shader or warp node in designer does. edit: the model is what informs the amount of change at a given pixel For Color its…
I didn't look at the model or read everything in detail, but I dimly recall having some problem with the normal map mixing mode in Substance in the past (replace vs. combine) and it not being really intuitive or behaving like expected. Might be worth doing a quick test if that changes anything.