Generation of Human Faces Using Generative Adversarial Network

GANs have proven to be a powerful deep-learning framework for creating realistic synthetic images, finding wide use across various tasks in computer vision. In this work, we introduce a GAN-driven method for the generation of human face images using a deep convolutional structure. We trained our model on 9,090 resized facial images to 128 × 128 pixels from the without-mask portion of the Face Mask Lite Dataset. In this adversarial setup, the Generator synthesizes facial images from random noise vectors, while the Discriminator distinguishes between real and generated samples. The main goal here is to build a lightweight, computation-friendly GAN that still yields visually convincing face images without resorting to heavy architectures. These experiments show that the model captures the essential facial features-symmetry, texture, and overall appearance-and generates a diverse set of synthetic faces. The evaluation combines the qualitative visual inspection of generated samples with quantitative analysis of Generator and Discriminator loss trends. The results point toward stable training, realistic face generation, and a preference for architectural simplicity.