Abstract: Aerodynamic design plays a vital role in rocket design. To expedite the design process of rocket aerodynamic shapes and minimize the tedious iterative optimization, we propose a generative rapid inverse design method for rocket aerodynamic shape design. This method uses image-based geometric characterization of rocket aerodynamic shapes and integrates three main components: a rocket aerodynamic shape generation model based on a conditional diffusion model, a rocket aerodynamic performance rapid prediction model based on a convolutional neural network, and a rocket design parameter discrimination model. This method operates in three consecutive steps. First, it uses the aerodynamic shape generation model and aerodynamic performance indicators as input to efficiently generate a large number of imaged rocket aerodynamic shape schemes. Second, the prediction model is used to select the design scheme that meets the input indicators. Third, the selected scheme is verified after extracting parameters through the design parameter discrimination model or converted into a three-dimensional shape. The model is validated using a data set containing 165,000 sets of aerodynamic shapes and corresponding aerodynamic performance for a typical axisymmetric rocket. Taking the axial force coefficient C_A and the normal force coefficient C_N under typical working conditions as input, the model only takes about 30 seconds to generate 100 rocket aerodynamic shape images. The optimal design exhibits remarkable accuracy, with deviations of only 0.0827% and 0.7124% for the axial and normal force coefficients, respectively. The results demonstrate the potential of the proposed method for rocket aerodynamic shape design. In addition, the influence of the cross-attention mechanism on the model is analyzed, providing valuable insights for the future model optimization.