This thesis explores the potential of a Constrained Deep Convolutional Generative AdversarialNetworks (CDCGAN) to reconstruct detailed facial images from feature vectors. The feature vectors in this project captures critical facial characteristics such as geometry, texture, and key landmarks, which the model seeks to convert back into corresponding images. The study focuses on the efficacy of a CDCGAN in handling complex data transformation, with potential applications inareas like digital forensics. The results showed potential and challenges of CDCGANs in generating high-fidelity facial reconstructions from compressed data.