Here’s a video model signal worth watching: OmniShow is trying to make human-object video less like prompt roulette. It unifies text, reference image, audio, and pose in one HOI pipeline. That…