A simple way to get better resolutions is to use the “progressive” approach. Basically, you train a 32x32 GAN, then you add another up-sampling layer and train a bit more, producing 64x64 results. Repeating the procedure, you will get 128x128 imagens, and so on.

The idea is simple, works virtually with any model, and also improves the quality. You may also decrease the learning rate for each specific resolution to get better results.