Batik is a traditional handicraft of ethnic minorities in southwestern China’s Guizhou province. It carries distinctive features, and inherits profound artistic and cultural values. However, the empirical design model of batik patterns is not fast or diverse enough to meet the highly personalized market demands. Moreover, it is difficult to parametrize the complex ethnic patterns with mathematical models. Hence, the design of batik patterns cannot satisfy the visual cognitive needs of consumers, which are constantly changing, fuzzy, and personalized. In other words, the visual cognitive needs of consumers have not been fully considered in the parametrization and design of batik patterns. To solve the problem, this paper puts forward a generative design method for batik patterns based on shape grammar, and relies on the artificial neural network (ANN) to model the nonlinear mapping between design parameters and visual cognitive image (VCI) values. The three-layer ANN model was optimized with the genetic algorithm (GA) to generate new patterns, and optimize the composition of their VCIs. The workflow and effectiveness of the proposed method were explained and verified through experiments on the generative design of bronze drum patterns, which have rich ethnic and religious connotations. The experimental results show that our method can predict the VCI values of the composition, provide the optimal parameter solution of the VCI values, and decompile the conceptual prototype from the pattern composition that best meets the visual cognitive needs of consumers.