The document discusses using generative adversarial networks (GANs) for text-to-image generation. GANs involve two neural networks, a generator and discriminator, that compete against each other. The generator generates images from text descriptions, while the discriminator tries to distinguish real images from generated ones. The document outlines the network architecture, literature review on GAN improvements, methodology used which involves training the GAN on a dataset to generate high resolution images from low resolution inputs conditioned on text.