For the analysis of Imaging Air Cherenkov Telescopes (IACTs) data, numerous air shower simulations are needed to derive the instrument's response.A process that is both computationally intensive and often requires repetition under varying observation conditions.Generative models based on deep neural networks offer an ultra-fast and more efficient alternative, significantly accelerating simulation times while compactly storing vast simulation libraries.Previous works focused on the generation of gamma showers; however, mostly proton showers need to be simulated for a good background description that features larger fluctuations, making their generation significantly more challenging. In this study, we employ diffusion models to generate proton showers for an IACT with nearly 2,000 pixels.Using simulations from the H.E.S.S. experiment, we assess the quality of the generated images via low-level observables and established shower shape parameters.While the generated images demonstrate high-quality low-level properties, further refinement is needed in modeling distinct shower shapes.