This project investigates multimodal generative artificial intelligence for sensing and localization in future wireless systems. The main objective is to develop learning-based methods that can jointly exploit heterogeneous sensing information, such as wireless signals, visual observations, environmental context, and other available sensor modalities, to improve localization accuracy, robustness, and generalization. Generative AI models will be explored for learning compact multimodal representations, modeling complex relationships between observations and physical environments, and generating or reconstructing useful sensing information when measurements are noisy, incomplete, or unavailable. Particular attention will be given to foundation-model-inspired architectures and multimodal representation learning that can transfer knowledge across different environments, sensing configurations, and localization tasks. The project will evaluate these methods using simulated and/or experimental datasets and compare them with conventional model-based and discriminative machine-learning approaches. The expected outcome is a better understanding of how multimodal generative AI can enhance wireless sensing and localization, especially in challenging and previously unseen environments. The developed methods may contribute to future applications in integrated sensing and communication, positioning, digital twins, and intelligent wireless systems.