Generating Human Faces from Feature Descriptions Using DCGANs

This research presents a novel text-to-face generation system using Deep Convolutional Generative Adversarial Networks (DCGANs) trained on the CelebA dataset, which comprises over 200,000 images with diverse facial attributes. The proposed model integrates BERT embeddings to enhance the semantic alignment between textual descriptions and generated images. Comparative analysis reveals that the DCGAN achieved an Inception Score of 2.840, a Frechet Inception Distance of 87.146, and a Clean FID of 87.580 performance against SAGAN and DFGAN models. These results highlight the effectiveness of our approach in generating realistic and contextually accurate facial images from text prompts.