Abstract: To improve the accuracy and efficiency of identifying carbonate fault-karst reservoirs, we developed a prediction model based on artificial intelligence (AI) technology through four steps: Gnerating a fault prediction model, generating a karst cave prediction model, data augmentation, and constructing an improved high-resolution convolutional neural network. The comparative test shows that the AI-based prediction outperforms conventional amplitude curvature attributes in processing time reduction and the ability to identify more dissolved caves, with its predictions consistent with seismic responses, well data, and established interpretation schemes. We use this method with supporting techniques to establish an AI-based fault-karst prediction workflow in the field application in the Tarim Basin, which mainly includes four steps: AI-based fault prediction, preprocessing of original seismic data using a seismic background modeling technique, AI-based karst prediction, and principal component-based fusion of AI-based fault and karst attributes to characterize the fault-karst structure. According to the prediction results, five wells were deployed, with test production exceeding twice the individual-well average daily output in the area of interest. This method and workflow can be used for the evaluation of fault-karst reservoirs in similar areas, providing technical support for carbonate exploration and development.