How to generate disorderly variable data with GAN when small data set


I’m training Conditional GAN with Small dataset
and i’m wondering if there is i can do for generating disorderly variable data with GAN, which is very light network (MLP based 3 layers)
I mean, once the GAN is trained with the limited number of samples,
I can select the random z values for generating the fake data
but the generated ones are quite easily overfitted but I want for the generator to produce randomly and disorderly distributed fake data.

Please help!