In this paper, we introduce the application of machine learning and deep learning in the field of soil physics. We then propose using generative AI for saturated and unsaturated flow analysis for artificial embankments such as fill dams and dikes. The generative AI method employed is a generative adversarial network (GAN) model incorporating a U-net. The relationship between hydraulic conductivity distribution and pressure head distribution in a steady state inside the dike is considered as an image-to-image translation problem. For the forward problem of estimating the distribution of pressure head, the accuracy was close to that of the conventional numerical method. On the other hand, for the inverse problem of estimating the distribution of hydraulic conductivity, the estimation accuracy was low. The estimation method using generative AI for the interior of the dike can serve as a means of achieving a digital twin of the dike because the data can be analyzed immediately. This approach is expected to be useful for real time evaluation and anomaly detection in dikes.