This study proposes an improved GAN-based surrogate model for rapid urban wind environment prediction. By integrating multi-dimensional geometric and physical information—such as multi-height signed distance fields, building and airspace boundaries, and wind direction—into unified multi-channel .npz format tensors, the model simultaneously predicts multiple wind variables (Ux, Uy, Uz, p) and more accurately captures complex 3D features. Compared to previous models, the proposed method achieves lower mean squared error (MSE) and higher structural similarity index (SSIM) across both simple and complex geometries. Despite increased input and output dimensionality, the model maintains fast inference speeds, requiring only ~4 seconds per case, significantly outperforming traditional CFD simulations. All training data are generated in OpenFOAM, ensuring consistency and reproducibility. Moreover, the model infers spatial scaling implicitly through SDFs, enhancing generalizability across diverse urban morphologies. Overall, this work provides an efficient and scalable approach for real-time urban wind flow prediction.