<p>The dynamic prediction of landslide deformation constitutes an essential component of landslide early warning systems. The implementation challenges for regional landslide risk management primarily stem from three aspects: the elevated costs of in-situ monitoring, the Limited temporal resolution of satellite observations, and the morphological diversity of slope failures. In this study, an InSAR-driven prediction framework is proposed to predict multi-type landslide deformation characteristics, including displacement and failure time. The inverse function of the Linear logarithmic transform is initially established to correct multi-type landslide displacements from SBAS-InSAR by considering slope gradient, and the results achieve strong agreement between the corrected and monitored data. Empirical mode decomposition, fitting analysis, deep learning of gated recurrent unit, and the modified inverse velocity method are integrated to predict the displacement and failure time of four types of landslides in Zigui, Three Gorges Area. Considering the impact of precipitation and reservoir water level, the average accuracy of displacement prediction for four types of landslides reaches 85%, while failure time prediction errors decreased from an average of 23.85 to 3.07 days. Field validation demonstrated framework effectiveness, particularly in instrumentally underserved areas. Although the uncertainty in the prediction of failure time for wading landslides is slightly higher, the framework could still provide an alternative way to obtain landslide deformation information and predict its development trend dynamically in areas not covered by specialized monitoring or SAR images over a certain period. This methodology enhances low-cost regional landslide early warning capabilities through optimized InSAR-data utilization and a hybrid deep learning model.</p>