PROMPT RADAR / VIDEO视频叙事
She just stepped on his nose to get past him. Then her foot decided the
She just stepped on his nose to get past him. Then her foot decided the best route was through his mouth. Same glitch as last time: two objects end up too close together in frame, and the model loses track of where one e
@voyzlab 245 热度 1.7 万 浏览
编辑先看
它为什么值得进入灵感库
先别急着照抄整段 Prompt：真正有用的是主体、空间和镜头之间的约束关系。 把一段描述拆成起始状态、动作变化、镜头响应和结束画面，连续性比形容词更重要。
01 / 原始方法
动作与镜头 Prompt
She just stepped on his nose to get past him. Then her foot decided the best route was through his mouth. Same glitch as last time: two objects end up too close together in frame, and the model loses track of where one ends and the other begins. Last time it was a whole person sinking into a mouth. This time, one foot was enough. How to avoid bugs like this when working with AI video: LOCK - lock each character and object to its own reference using Nano Banana or Midjourney Omni Reference SEPARATE - pick a model with strong multi-subject tracking - Kling 3.0 or Seedance 2.0 DISTANCE - at the reference stage, keep contact points (hand-to-face, foot-to-head) a bit further apart than in the final scene CHECK - specifically re-watch the moments where objects touch, not just the whole scene She just wanted to climb over his head. Instead, she became part of breakfast. Tomorrow I'll break down something exclusive.
02 / 复刻路径
不要一次改完所有变量
用 Midjourney 复刻时，建议先单独生成稳定首帧，再把动作拆成“开始、变化、收束”三段；每段只保留一个主要运动。
- 主体：She just stepped on his nose to get past him. Then her foot decided the best route was through his mouth. Same glitch as last time: two objects end up too close together in frame, and the model loses track of where one e
- 视觉方向：实时 AI 视频案例；查看原帖媒体与作者说明
- 镜头与构图：按原帖镜头流程执行；查看原帖媒体与工作流
- 交付约束：auto，保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束
保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
模型建议
Midjourney / Seedance / Kling / Veo。先用原始 Prompt 建立基准结果，再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。
来源与编辑原则