TL;DR of the paper, generated automatically.

Research paper proposing a video model that reasons about how events change a scene.

• Problem: video models render plausible frames but miss consequences, like a person fleeing a beast.
• Method: a vision-language model tracks world state while a video renderer draws it.
• Results: the paper says it beats recent models on event-driven causal evolution, keeping visual quality.