PROMPT RADAR / VIDEO视频叙事
7 open source skills that run a full AI video production pipeline end to
7 open source skills that run a full AI video production pipeline end to end, built for Seedance 2.5 and working in Claude Code, Codex, Hermes or any harness. → /setup writes your model stack once so every skill reads th
@RoundtableSpace 474 热度 4.0 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
7 open source skills that run a full AI video production pipeline end to end, built for Seedance 2.5 and working in Claude Code, Codex, Hermes or any harness. → /setup writes your model stack once so every skill reads the same config → /studio-init scaffolds the entire studio file tree from one question → /film-breakdown writes a 22-field card for every shot in your script → /reference-board locks references into a visual bible with a ban list → /asset-passport writes the exhaustive descriptor every later prompt quotes word for word → /stress-test flips assets to locked only at 10 out of 10 repeatability → /shot-prompt refuses to run until everything in frame is locked, then writes the 15-block prompt
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：7 open source skills that run a full AI video production pipeline end to end, built for Seedance 2.5 and working in Claude Code, Codex, Hermes or any harness. → /setup writes your model stack once so every skill reads th
- 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