While you could train your model after generating one image, it is extremely inefficient and leads to less stable training. In GANs, and in machine learning in general, you will process multiple images per training step. These are called batches.