Analyzing microstructures plays a crucial role in both metallurgical science and manufacturing, driving considerable investment in the preparation and imaging of structured materials. In this study, we introduce a deep learning approach utilizing conditional diffusion models to synthesize electron backscattered diffraction (EBSD) misorientation maps of quenched and tempered steel based on light optical microscopy (LOM) images. By harnessing the affordability and ease of producing LOM micrographs to generate high-quality EBSD misorientation maps, our goal is to streamline the characterization phase in steel production. This research is backed by our unique dataset, collected by our German collaborations, comprising aligned LOM and EBSD images captured from identical sample locations and magnifications. We demonstrate the potential of diffusion models in the field of materials science imaging by reconstructing EBSD misorientations from LOM images of complex, multiphase steel. Our findings suggest that while diffusion models can generate plausible and internally coherent EBSD misorientation maps, their absolute misorientation values may lack precision. Additionally, expanding to the use of polarized LOM is an active and promising area of research, offering the potential to further enhance the microstructural information contained within the optical data.