Astronomy is constantly discovering never-befor observed phenomenon. The idea that you can simulate realistic images of anything requires you to have sufficient knowledge of reality, and astronomy keeps showing us that we don't have that.
The only way I can see this being helpful is to train algorithms for what is already known and can be safely filtered out, making it easier to detect new observations