Ah, I understand what you are saying. The instabilities could be explained by the batches sampled during those training days and the generator's input. Training

Ah, I understand what you are saying. The instabilities could be explained by the batches sampled during those training days and the generator's input. Training a GAN is not very straightforward and even minor changes in batch sampling could produce vastly different generated images.