In my PhD research, I probed into the synergistic relationship between visual recognition and synthesis models, illustrating how these two domains can mutually enhance each other. Currently, at Adobe, I am advancing this exploration by leveraging recognition models to augment visual synthesis and editing. This effort involves innovative techniques such as automatic artifact localization and refinement, all aimed at enhancing the overall quality of visual generation.