In the second part of our impervious surface series, we demonstrate how freely available true orthophotos can be transformed into a detailed, high-resolution land cover classification using modern AI architectures. Building on our flexible AI framework, we provide an overview of current AI developments and demonstrate their added value in the semantic segmentation of urban surfaces. The goal is to reliably derive as many relevant surface classes as possible – creating a robust data foundation for municipal applications, ranging from climate resilience to infrastructure planning.