Since decades, researcher are investigating the modeling of the fracture morphology of glasses—especially for prestressed glasses. As an alternative to existing mechanics-based or explicit statistical approaches, we propose neural implicit modeling via Neural Cellular Automata (NCA) to simulate microstructure development during the fracture process in pre-stressed glasses. Based on convolutional neural network, NCA can learn essential fracture features, such as preferred growth direction and geometric features of the fracture particles. The proposed NCA are more accurate than the “BREAK” method towards texture and geometrical features as well as orders of magnitude faster than the conventional Phase-Field or Finite-Element-based models. While this study employs images of fractured glass panes as training data, NCA can also be trained based on any microstructural simulation data or a mix of synthetic and actual data.