音频驱动视频时，提示词描述角色和镜头意图，音频负责节奏与口型参考。先用短音频验证同步，再增加镜头复杂度，避免把音频节拍、人物动作和场景变化一次塞进模型。 For audio-driven video, use the prompt for character and camera intent while the audio provides timing and lip-sync cues. Validate with a short clip first, then add camera complexity instead of changing every variable at once.