Converting old videos to modern HDR with AI

Most existing videos are in standard dynamic range (SDR), but HDR displays are becoming common. This work solves the SDR-to-HDR conversion problem using a two-part deep learning approach: a Multi-Exposure Video Model (MEVM) generates multiple exposure variants from a single SDR input, then a Video Merging Model (VMM) combines them into a high-quality HDR sequence that preserves shadow and highlight detail. User studies and quantitative evaluation confirm the method works on casual consumer videos and archival film. The framework integrates with existing generative video models and code is available.