This minisymposium addresses generative methodologies for the automated creation of physics-based models and digital twins, with applications in early-stage engineering design, simulation-driven development, and process automation. Such methods accelerate model development by leveraging diverse information sources, including (un)structured data, symbolic or mathematical descriptions, heterogeneous sensor or process data, and domain-specific knowledge, while maintaining physical consistency, interpretability, and ideally treating (epistemic) uncertainty. The motivation for this minisymposium stems from the growing need to develop models in settings where traditional manual modeling is slow and unflexible. Generative approaches can integrate knowledge from physics, data, and prior models into unified, adaptive representations for fast and reliable model generation. A central theme is AI-driven generative models' rigorous verification and validation (V&V) to ensure their reliability in industrially relevant and safety-critical contexts. Verification should confirm physical consistency, numerical robustness, and the absence of hidden failure modes, while validation -guided by frameworks such as VV10 or VV20- should demonstrate predictive capability against trusted experimental or high-fidelity simulation benchmarks. We particularly encourage methods that integrate V&V and explainable AI into the generative process, making model behavior transparent, clarifying the role of physical constraints and data, and enabling continuous assessment of fidelity and reliability throughout the model lifecycle. Relevant application areas range from medicine and biomechanics to industrial systems, including discrete and continuous manufacturing, large-scale multi-physics simulations (e.g., FEM or CFD), and scenarios aligned with industry standards such as FMI/FMU-based co-simulation. Contributions may also address integration with legacy simulation environments, such as industrial software, system simulation platforms, operational digital twins, or multi-physics codes, to enhance the interoperability and practical deployment of generative methods. Technologies of interest include physics-informed and probabilistic machine learning, diffusion models, Bayesian model discovery, physics-constrained optimization, symbolic and (knowledge) graph-based analysis, large language models with integrated reasoning, causal modeling, and uncertainty-aware inference.