Generative models offer promising applications for inverse problems in physics, fast surrogate models, and the search for new physics. Additionally, generative models have recently been proposed for advancing our understanding of nucleons and nuclei when compared with recent advances in nuclear theory. Notably, score-based generative diffusion models have demonstrated the ability to produce high-fidelity, accurate samples and have addressed slow generation times through distillation. This works expands on previous generative diffusion models for collider data in three distinct ways. First, a model is trained on simulations for the Electron-Ion Collider to generate the entire electron-proton collision. It accurately generates all particle species and their complete kinematic information, conditioned only on the beam type and energy. Second, it is the first work to adapt a pre-existing foundation model developed for high energy physics called \textsc{OmniLearn}. This points to a transition away from designing new models from scratch for each study, to a future of re-using foundation models with wide-ranging capabilities. Third, many diffusion model implementations for high energy physics rely on image based techniques, which fall short due to the sparsity of the data. This issue is addressed by using point clouds and a Point Edge Transformer (PET) architecture. This work marks a significant step forward in the application of generative models to collider physics.