Possible solution for transparent generation.

Can we paint over an object, pull a mask off of an object, or have the A.I. pick objects based on a grid picker of the image recognition, then create layer masks? Then assign GPUs to certain layers to render with different prompts.