This paper presents a deep-learning framework, Multi-load Generative Adversarial Network (MultiLoad-GAN), for generating a group of load profiles in one shot. The main contribution of MultiLoad-GAN is the capture of spatial-temporal correlations among a group of loads to enable the generation of realistic synthetic load profiles in large quantity for meeting the emerging need in distribution system planning. The novelty and uniqueness of the MultiLoad-GAN framework are three-fold. First, it generates a group of load profiles bearing realistic spatial-temporal correlations in one shot. Second, two complementary metrics for evaluating realisticness of generated load profiles are developed: statistics metrics based on domain knowledge and a deep-learning classifier for comparing high-level features. Third, to tackle data scarcity, a novel iterative data augmentation mechanism is developed to generate training samples for enhancing the training of both the classifier and the MultiLoad-GAN model. Simulation results show that MultiLoad-GAN outperforms state-of-the-art approaches in realisticness, computational efficiency, and robustness. With little finetuning, the MultiLoad-GAN approach can be readily extended to generate a group of load or PV profiles for a feeder, a substation, or a service area.