This study trains a deep learning model for rapid and interactive design assessment in order to overcome the time constraints of conventional hydrodynamic simulations in urban flood prediction. A convolutional neural network (CNN) for flood depth and velocity estimation that directly uses topographical data and urban building layouts was trained and developed. The CNN eliminates the need for the hydrodynamic simulations by concentrating on these adjustable design characteristics, allowing architects and urban planners to quickly test different urban designs. A dataset of hydrodynamic simulation results of various urban morphologies was used to train and test the CNN model. Results indicate that the proposed CNN model reduces the computational time to seconds while maintaining an impressive accuracy comparable to the traditional hydrodynamic simulations. Real-time iteration of urban design options is made easier by this quick and precise performance, which enables practitioners to test and determine the building configurations and urban form that corresponds to a maximized flood resilience, and estimate flood hazard across various urban layout situations. Characterizing of flood depth and velocity surrounding constructed structures with great spatial precision potentially makes the proposed CNN model a practical and a real-life tool for flood-resilient urban planning. This study advances urban planning decision support tools by providing a methodology that makes it possible for flood resilience evaluation to be incorporated into the iterative design process.