Text-to-face generation, a sub-domain of text-to-image synthesis, holds significant promise for various research areas and applications, particularly in the realm of public safety. However, the progress in this field has been hindered by the scarcity of datasets, leading to limited research efforts. Most existing approaches for text-to-face generation rely on partially trained generative adversarial networks (GANs), where a pre-trained text encoder extracts semantic features from input sentences, which are then utilized to train the image decoder. In this study, we propose a fully trained GAN framework to generate realistic and natural images. Our approach simultaneously trains both the text encoder and the image decoder to improve accuracy and efficiency in generating images. Additionally, we contribute to the field by creating a novel dataset through the fusion of existing datasets such as LFW and CelebA, along with locally curated data, which is labeled based on defined classes. Through extensive experimentation, our fully trained GAN model has demonstrated superior performance in generating high-quality images based on input sentences. The visual results further validate the effectiveness of our approach in accurately generating facial images corresponding to the provided queries.