PROMPT RADAR / VIDEO视频叙事
"Don't post this" A creator with early Seedance 2.0 access sent me his r
"Don't post this" A creator with early Seedance 2.0 access sent me his render privately I'm posting it Two women fighting in a flooded bathroom. Same face, same dress, same reflection in the mirror through every cut. Ful
@cipgerx 750 热度 3.3 万 浏览
编辑先看
它为什么值得进入灵感库
它最有价值的部分,是把一个模糊想法推进到了可以被模型执行的程度。 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
"Don't post this" A creator with early Seedance 2.0 access sent me his render privately I'm posting it Two women fighting in a flooded bathroom. Same face, same dress, same reflection in the mirror through every cut. Full contact from frame one He didn't post this anywhere. Didn't make a tutorial. Just dropped the file in a DM and said "tell me this looks like AI" I couldn't Three documents. That's what separated his output from everyone else's failed generations → Character sheet: face, dress, hair, three angles, locked → Location sheet: bathroom, fluorescent light, water level, locked → Storyboard: every shot in order, locked before a single frame rendered She gets thrown into a door. Water sprays across tile. The camera holds her face. Not one feature drifts Production studios are quietly testing this on real pre-vis budgets. Nobody is posting about it publicly yet You're still rewriting prompts and wondering why your character changes face at frame twelve The prompt was never the problem. The three documents you never built were
02 / 复刻路径
不要一次改完所有变量
用 Seedance 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:"Don't post this" A creator with early Seedance 2.0 access sent me his render privately I'm posting it Two women fighting in a flooded bathroom. Same face, same dress, same reflection in the mirror through every cut. Ful
- 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
Seedance / Veo / Kling。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