Corrosion and cracking in additively manufactured (AM) metallic materials are strongly governed by complex polycrystalline microstructures, where multiscale spatiotemporal interactions control damage initiation and evolution. Accurate long-term prediction of such processes is critically important for reliability and lifetime assessment of AM components. Phase-field modeling has emerged as a powerful physics-based tool for simulating corrosion and crack evolution with high fidelity; however, its extremely high computational cost and sensitivity to parameters limit its use for rapid prediction and large-scale studies. To address this challenge, this work proposes a Generative AI method that accelerates phase-field simulations by learning microstructure-aware corrosion and crack evolution dynamics. The model takes crystallographic microstructure as input and generates the full spatiotemporal evolution of corrosion and cracking. A key novelty of the proposed framework is the incorporation of time-embedding conditioning, which enables adaptive time evolution and allows the model to accurately capture both corrosion-dominated and crack-dominated stages with non-uniform temporal dynamics. Several predictive schemes are systematically investigated, including one-step (one-frame-to-next), multi-frame (five-frames-to-next), and sequence-to-sequence (five-frames-to-five-frames) generation strategies, and their effects on stability, accuracy, and long-term prediction are analyzed. The trained model successfully reproduces critical physical phenomena, including corrosion front propagation, corrosion-to-crack transition, crack initiation sites, and crack path deflection and branching, while maintaining strong spatiotemporal consistency with phase-field results. The proposed framework predicts the full corrosion–crack evolution for unseen microstructures within seconds, achieving orders-of-magnitude acceleration compared to traditional phase-field simulations while preserving key physical mechanisms and time-dependent behavior. This work demonstrates a fast, reliable, and physics-consistent predictive tool for microstructure-driven corrosion and cracking in AM metals, offering significant potential for accelerated materials design and durability assessment.