Generative Adversarial Networks, which showcase a new form of generative modeling, function as a neural network generative model via deep learning, with two competing neural networks (i.e., the generator and the discriminator) competing against each other. During this interplay, the generator ‘masters’ the art of deceiving the discriminator into perceiving a given image as (superficially) genuine. This technique is frequently employed to create images derived from existing datasets. This paper investigates the utilization of Generative Adversarial Networks for the production of handwritten digital characters, using the MNIST dataset as a reference point. Moreover, we present an analysis of the relationships and differences between these three paradigms in training algorithms for generating handwritten digits from the MNIST dataset through the exploration of the Deep Convolutional Generative Adversarial Network and the Conditional Generative Adversarial Network models, providing an optimization strategy for improving the generation of handwritten numerals.