Residual stress is an important factor in manufacturing, including laser processing, and ignoring it can lead to product defects. This study presents a new deep learning approach to predict critical characteristics of laser-irradiated carbon steel, including thermal residual stress. The core of this study, a generative adversarial network (GAN), acts as a surrogate model, enabling more efficient utilization of computationally expensive simulations to generate the desired output. The GAN model is trained using data generated by numerical simulation. Three key laser processing parameters are taken as input: laser power, beam diameter, and irradiation time. Based on these inputs, the model predicts various properties of the laser-irradiated carbon steel throughout the process and after cooldown including residual stress. This deep learning model effectively captures the complex, non-linear relationships between laser parameters and process outputs. The model produces very fast results when compared to the traditional simulation, while the results are accurate. The model exhibits exceptional performance, achieving high R-squared values ranging from 0.975 to 0.999 across various predicted properties. The result demonstrates the effectiveness of the proposed deep learning approach for predicting laser processing outcomes in carbon steel. It facilitates efficient exploration of residual stress and melt pool evolution by enabling rapid evaluation of different laser parameter settings. Furthermore, this study emphasizes the benefits of utilizing a hybrid loss function for improved optimization of deep learning models.