Fake News Detection and Manipulation Reasoning via Large Vision-Language Models