Not sure I understand the question. This paper is about using diffusion models to reconstruct usable images from raw sensor data. The diffusion model in essence replaces the lens.