Methods for splitting a dataset efficiently

Question

I have a fairly large dataset of images. They have been taken by \'x\' number of photographers and each image falls into one of \'y\' themes. How would I go about making a train