The ability to create realistic facial images from text descriptions has become a topic of important interest lately because of its various uses in computer vision, entertainment, and virtual communication. This research introduces a novel approach, termed Bidirectional Deep Convolutional Generative Adversarial Networks (Bi-DCGAN), which integrates Deep Convolutional Generative Adversarial Networks (DCGAN) with Bidirectional Long Short-Term Memory (Bi-LSTM) networks to address the limitations of existing methods. Training the text encoder and image decoder simultaneously enables the model to effectively learn the relationship between textual descriptions and visual representations, utilizing a comprehensive dataset that pairs facial images with their respective text annotations. Furthermore, the study introduces a new dataset that merges the CelebA dataset with a locally collected dataset, boosting the quality of training. Experimental findings indicate that the fully trained DCGAN outperforms existing models in terms of image quality and alignment with input descriptions, demonstrating significant advancement in generating facial images that accurately reflect textual input.