Diffusion Models have recently been adapted to protein modeling, promising novel therapeutic design. However, current models lack physical principles. Trained on crystallized protein structures, they require computationally expensive post-processing via molecular dynamics for realistic applications. My project aims to bridge data-driven molecular generation with physics by integrating interatomic force fields directly into the generative pipeline. A physics-informed guidance scheme will steer the diffusion process efficiently toward meaningful protein structures. This approach will be extended for sampling multiple valid folded structures at finite temperatures, enabling extraction of key properties like binding free energy. Finally, I aim to investigate whether the physics-augmented diffusion models can be leveraged for simulating folding trajectories, in a hybrid approach that comes at much lower computational cost compared to classical molecular dynamics simulations. This research builds on my expertise in dynamic guidance for image generation, my training as an engineering physicist and further aligns with my supervisor's strategic direction in AI-based molecular modeling. The goal is to provide Flanders' antibody and nanobody industry with cutting-edge tools to enhance their protein engineering pipelines.