A segmentation model is trained using both. Through a continuous feedback loop, the system refines the images it creates based on how well they improve the model's learning.