Reliable fault detection, classification, and location prediction are crucial for maintaining the stability and resilience of electrical power systems. Conventional methods often struggle with nonlinear fault patterns, imbalanced datasets, and real-time requirements. This study introduces a Hybrid Generative AI model that combines Generative Adversarial Networks (GANs) for creating synthetic fault data, CNN-LSTM deep learning architectures for extracting spatiotemporal features, and tree-based ensemble models to refine decisions. The model was trained and tested on a representative dataset of simulated fault events in Maharashtra, India, covering four main fault types: Single Line-to-Ground (SLG), Line-to-Line (LL), Double Line-to-Ground (DLG), and Three-Phase (LLL) faults. Results show that the hybrid approach significantly improves detection, classification, and localization accuracy over traditional machine learning (ML) and standalone deep learning (DL) methods. The model achieves nearly perfect detection (Detection Accuracy = 98%, Detection F1 = 99%), high classification accuracy (87% with a macro-F1 of 83%), and strong fault localization with a normalized score of 82%. Overall, the results, visualized through radar analysis, confirm that the hybrid generative design effectively addresses the challenges of fault analysis for smart grid applications.