The use of generative deep learning models has been of interest in the high-energy physics community intending to develop a faster alternative to the compute-intensive Monte Carlo simulations. This work focuses on evaluating an ensemble of GANs on the task of electromagnetic calorimeter simulations. We demonstrate that the diversity of samples produced by a GAN model can be significantly improved by expanding the model into a multi-generator ensemble. We present a study comparing the single-GAN model and the ensemble model using both physics-inspired and artificial features.