WHAT-IF: Exploring Branching Narratives by Meta-Prompting Large Language Models