Fight Prompt Director — AI skill for Claude Code

Fight Prompt Director

中文名：动作导演Skill

一个面向 AI 视频生成的动作戏提示词编排 Skill。它把角色、场景、图片、首尾帧或分镜素材整理成可直接使用的视频提示词，重点解决动作因果、角色差异、空间方向、镜头职责和跨镜连续性。

English summary: A bilingual prompt-directing skill for AI-generated fight and action scenes. It turns text, images, keyframes, or storyboard references into executable video prompts with readable action, spatial continuity, camera intent, and character-specific movement.

核心能力

• 为每名角色设计不同的优势距离、移动方式、攻击习惯和反制策略
• 将每轮交锋写成完整因果链，覆盖发起、接触、回应、位移和下一拍接口
• 支持近战、远攻、混合距离、人兽战、单人演武和风格化材质动作
• 为 Seedance 生成连续的粗时间线
• 为 MiniMax-H3 生成首尾衔接的精确时间码
• 输出高强度、慢节奏和中间型三套动作方案
• 提供标准版与极速版两档速度，极速版让三套强度方案同时变快
• 管理角色数量、武器、弹药、运动方向和场景损伤的跨镜连续性
• 提供动作、镜头和参考素材三类问题的单变量排错方法

What it does

• Builds a distinct movement profile for every character
• Converts exchanges into visible cause-and-effect action
• Supports melee, ranged combat, mixed-distance encounters, creature fights, solo forms, and crafted visual styles
• Uses coarse continuous time blocks for Seedance
• Uses contiguous precise timecodes for MiniMax-H3
• Produces high-intensity, measured, and hybrid versions
• Offers a standard and an ultra-fast speed setting, where ultra-fast speeds up all three versions
• Tracks position, facing, weapons, projectiles, motion, and environmental damage across cuts
• Diagnoses failures by changing one variable group at a time

为什么动作提示词需要导演结构

动作视频同时依赖角色一致性、肢体关系、接触位置、屏幕方向、镜头衔接、道具稳定和环境记忆。只有“快速打斗”之类的概括性描述时，模型往往会缩减动作、隐藏接触或重置空间。

这个 Skill 会把动作拆成能够观察的关系：

1. 谁先发起
2. 动作指向哪里
3. 接触、格挡或落空
4. 对方怎样回应
5. 双方位置怎样变化
6. 什么状态让下一拍成立

它同时要求重要命中交代加速、接触、受力、位移和环境反馈，让力量通过画面结果呈现。

Why director-style prompting helps

Action scenes ask a video model to maintain identity, body logic, contact, screen direction, props, camera continuity, and environmental memory at the same time. Vague intensity words often lead to reduced motion, hidden contact, or spatial resets.

This skill turns action into observable relationships. Each exchange identifies the initiator, target, contact outcome, reaction, disp