cinematic-narrative

MuSS introduces a large-scale dataset and benchmark for multi-shot subject-to-video generation, addressing narrative logic and copy-paste issues in cinematic storytelling.