Abstract: Influenced by complex topography, Southwest China (SWC) is prone to floods and frequent secondary geological hazards induced by heavy rainfall during the rainy season. Thus, effective climate prediction is of great significance for disaster prevention and mitigation. This study integrates machine learning (ML), linear regression (LR), and the interannual increment method to develop a real-time monthly precipitation prediction model for the rainy season (May–September) over SWC. Summer precipitation is subsequently derived from monthly results. Firstly, empirical orthogonal function (EOF) analysis is applied to extract modes of monthly precipitation. Based on preceding reanalysis data and simultaneous predictions from the second version of Climate Forecast System (CFSv2) released in February, a pool of predictors is established through image recognition techniques and subsequently refined based on the information-flow method and their relationships with contemporaneous circulation anomalies. Then, multiple predictor sets are constructed by setting different thresholds for key factors affecting statistical forecast skill, such as correlation and independence. For each month, hindcasts are conducted using four machine learning models (Prophet, TabPFN, Support Vector Regression, and Random Forest) as well as Linear Regression (LR). Based on hindcast performance, the skillful predictor sets and models are selected. By integrating the advantages of different models, a robust multi-model ensemble (MME) prediction model is developed. Hindcast results for 2011–2022 show that the MME effectively improves rainy-season precipitation prediction skill in SWC. The multi-year mean spatial anomaly correlation coefficients (ACC) for May–September and summer average precipitation are respectively 0.28, 0.21, 0.23, 0.25, 0.32, and 0.19, passing the 95% confidence level, and PS scores are 81, 78, 80, 78, 79, and 77. These results represent average improvements of 0.24 in ACC and 15 points in PS over CFSv2. The MME has been applied to real-time predictions for 2023–2025, providing a novel and effective approach for improving rainy season precipitation prediction in SWC.