Historical scientific archives often contain observations of lasting value that were acquired with lower-resolution instruments than those available today. Solar magnetograms provide a motivating example: the Michelson Doppler Imager (MDI) aboard the Solar and Heliospheric Observatory (SOHO) recorded decades of solar activity before the higher-resolution Helioseismic and Magnetic Imager (HMI) became available. We develop a generative framework that first learns how modern observations are transformed by an older instrument, uses this transformation to construct paired training data, and then trains a conditional diffusion model to reverse the loss of resolution. A scalable multiscale U-Net and DPM•  sampler generate multiple modern-quality reconstructions from each historical observation. More broadly, this provides a general strategy for connecting scientific archives across generations of instruments.