Abstract: The Chinese seismic instrument intensity prediction is crucial for earthquake early warning (EEW) and hazard mitigation in China, but traditional methods suffer from issues such as insufficient accuracy and insufficient fusion of multi-source data. This study aims to construct a multimodal deep learning model, explore its feasibility for predicting seismic instrument intensity in China, and improve the accuracy and robustness of instrument intensity prediction for EEW. A Multimodal Chinese Instrument Intensity prediction Network (MCIINet) is proposed, which is trained and tested by the seismic events recorded by China Earthquake Networks Center. Experiments have shown that on the test dataset, compared to the baseline model at 3 s after P-wave triggering, MCIINet reduced MAE and RMSE of instrument intensity prediction by 9.03% and 8.67%, respectively, and improved R² and accuracy by 9.10% and 2.51%, respectively. MCIINet has effectively improved the accuracy of intensity prediction through multimodal deep feature fusion, verifying the feasibility of multimodal deep learning for seismic instrument intensity prediction in China, and providing technical support for instrument intensity prediction in EEW.