Convolutional Neural Networks (CNNs) prove effective in workflows that facilitate the conversion of raster images of unreinforced masonry buildings into models for computational analyses leveraging the Discrete Element Method (DEM). CNNs capable of performing object detection and instance segmentation enable the replication of masonry units and pre-existing damage by detecting and delineating bricks and cracks. For this study, two YOLOv11 segmentation models are trained to detect and provide polygon masks of bricks and cracks found on orthophotos of a masonry façade. The masks are then used by Python and MATLAB-based algorithms to automatically generate discrete blocks following the simplified micro-modelling approach. The detected cracks are explicitly embedded into the discontinuous medium, applying dry joint bond properties to the corresponding mortar joints and cracks within bricks. The case study façade is simulated under lateral loads both with and without pre-existing cracks to prove the adopted concept and demonstrate the necessity of considering existing flaws/cracks in the structural analysis.