<p>Bayesian model updating (BMU) is essential in developing digital twins (DTs) by providing a systematic framework for integrating physical observations or measured data, with uncertainty quantified, with finite element (FE) models. As a widely adopted technique for creating physical-to-virtual (P2V) twinning, BMU enhances the accuracy of digital FE models in representing the (probabilistic) true state of physical systems, thereby supporting risk-informed decision-making in DTs. Despite its widespread adoption in various applications, BMU faces several fundamental challenges, namely the reliance on frequently intractable likelihood functions, the high computational cost associated with model evaluations, and the difficulty of accurately approximating multi-modal posterior distributions. These limitations often hinder BMU application in large-scale and/or complex systems. Recent advances in generative artificial intelligence (AI) offer new opportunities to overcome these limitations. By learning underlying complex data distributions, generative AI models can either accelerate traditional likelihood-based methods such as Markov chain Monte Carlo (MCMC) or enable likelihood-free inference. In this paper, we provide a comprehensive review and tutorial on how generative AI can be leveraged to enable for effective BMU in engineering applications. This review paper aims to provide a structured and comprehensive guide for understanding, evaluating, and applying generative AI techniques to improve the efficiency, scalability, and accuracy of BMU. We begin by presenting traditional BMU methods and outlining the motivation for integrating generative AI models. Then, we introduce the mathematical foundations and operational principles of four major generative AI modeling paradigms–variational autoencoders (VAEs), generative adversarial networks (GANs), normalizing flows (NFs), and diffusion models (DMs)—in a tutorial style accessible to a broad research community. Their capabilities in posterior inference are evaluated both qualitatively and quantitatively through a benchmark case study involving a linear elastic multi-degree-of-freedom (MDOF) system. Furthermore, a comparative performance analysis of these generative AI models is presented, highlighting their respective strengths and limitations. Finally, we discuss current challenges, practical considerations, and promising directions for future research in applying generative AI to advance BMU.</p>