Ensuring the security of face authentication systems is crucial, and Face Anti-Spoofing System (FAS) play a key role in defending against spoofing threats. Depth-supervised learning has proven effective in FAS, utilizing depth maps as auxiliary features due to their computational simplicity. However, existing methods often struggle to generalize effectively in intricate environments and counter unknown attacks. To address this challenge, our work introduces a novel GAN-based architecture for FAS. To enhance generalization, we introduce Multi-Scale Retinex with Color Restoration (MSRCR) images alongside RGB, and apply the Convolutional Block Attention Module (CBAM) mechanism within the generator framework to highlight salient features. The classifier is trained using a latent variable encompassing depth information, improving generalization across diverse environmental conditions, including variations in illumination and background. Experimental results demonstrate the effectiveness of our approach, outperforming other methods on multiple datasets including CASIA-FASD, MSU-MFSD, OULU-NPU and Replay-Attack for both intra-dataset and cross-dataset testing between Replay-Attack and CASIA-FASD datasets.