Predicting the damage behavior of 3D woven composites subjected to impact loading has long been a key focus of research. Finite element (FE) simulation is commonly employed to address such issue. However, it demands significant computational resources and is also time-consuming. To facilitate rapid prediction of the damage morphology in 3D angle-interlock woven composites (3DAWCs) under impact loading, this study implements FE simulation methods to generate a dataset of damage morphology and develops a predictor-decoder deep learning model that incorporates attention mechanism. This model is capable of providing accurate, real-time predictions of meso-scale damage and constituent damage. The successful development of this model demonstrates the feasibility of using deep learning frameworks for damage prediction in woven composites. It also offers a potential solution to the challenge of damage prediction for related textile materials and supports real-time material monitoring and structural design optimization.