街头小吃样本把摊位放在近景，把行人和建筑放到中远景，形成前中后景层次。生成提示词先写空间层级，再写人物动作，模型更容易保持主体和背景的相对关系。 A street-food sample layers the stall in foreground, people in midground, and buildings in the distance. Describe spatial layers before actions in a generation prompt to preserve subject-background relationships.
#OpenVid-1M#depth layers#candid
来源:OpenVid-1M · Street Scene Depth许可: CC BY 4.0 dataset card; research/non-commercial notice and upstream licenses apply