把 CogVideoX 提示词写成单条可观察的动作链：主体做什么、动作如何变化、镜头如何响应。减少同一时间发生的并行动作，模型更容易保持时间连续性。 Write a single observable action chain for CogVideoX: what the subject does, how the action changes, and how the camera responds. Fewer simultaneous actions usually improve temporal continuity.