Tips to creating a sequential GAN image generator

I would like to train a GANs model that can generate 1 output image from a set of 3 images.

  • Do you have any tips on this?

  • Will the GPU capacity needed to train this be really high? I just requested 1 P optimized instance on AWS to use a p2.xlarge instance. Will that be enough or do I need a p3.2xlarge ?