PROMPT RECORD图像记录

然后我顺着分镜清单逐行交给 Claude Opus,让它为每一行构建 Seedance 2.0 提示词。 它把光照规则和连续性守得比凌晨两点的我还紧。 唯一一条它绝对不能打破的规则是:只用具象...

中文说明

然后我顺着分镜清单逐行交给 Claude Opus,让它为每一行构建 Seedance 2.0 提示词。 它把光照规则和连续性守得比凌晨两点的我还紧。 唯一一条它绝对不能打破的规则是:只用具象光源。光线必须来自画面里的某个物体。一旦你加上电影级的补光,它立刻又像 AI 生成的。 我遵循的 Seedance 2.0 提示词结构: - 该镜头的参考(角色设定表和场景空镜),要求完全对齐。 - 来自你走位方案的空间布局,以锁住位置。 - 每个镜头的运镜、运动和动作,表情要在声音出现之前先切换到新情绪。 - 一个锁定的风格块:具象光、柔和粉彩色调、真实物理、没有身份漂移。 - 时长和画幅比放在最后。

原始 Prompt

Then I go down the shot list and hand each row to Claude Opus to build the Seedance 2.0 prompt.

It holds the lighting rules and continuity tighter than I do at 2 a.m.

The one rule it can never break is practicals only. Light has to come from something in the frame. The second you add movie fill, it reads as AI again.

The structure I follow for Seedance 2.0 prompts:

- The shot's references (character sheets and location plate), told to match exactly.
- The spatial layout from your blocking, so positions stay locked.
- Camera, move, and action per shot, with the face flipping to the new emotion just before the voice.
- A locked style block: practical lighting, pastel palette, real physics, no identity drift.
- Duration and aspect ratio at the end.