360 Secure SkillHub
360 安全技能中心

zmg-srew

核心功能是将新闻主题、活动策划、短视频、公益广告等各类创意,转化为专业可拍摄的分镜脚本,包含画面、景别、旁白、采访设计、素材清单等细节,适配制作人员及AI视频工具。典型场景为内部Demo制作,输入可提供故事设定、时长格式、受众风格等信息,输出结构化Markdown脚本,含创作定位、分镜表等模块,确保制作清晰可控。

jkbwfi
jkbwfi
热度 26.62
最近更新 2026-05-14 10:51:47
分类:创作依赖:纯提示词应用场景:故事创作 / 视频编辑 / 品牌素材 / 社交媒体

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add stoe
Metadata
name
zmg-storyboard-generator
description
Create shootable storyboard scripts for ZMGClaw internal demos. Use when the user asks to turn a news topic, campaign concept, documentary idea, short-video idea, public-service message, brand story, or creative outline into a scene-by-scene script with shots, narration, visuals, interview beats, rhythm, and production notes.
version
1.0.0
user-invocable
true

ZMG分镜脚本生成

Purpose

Transform an idea into a production-facing storyboard script. Optimize for clarity: directors, editors, reporters, camera operators, and AI video/image tools should understand what to create.

Inputs

Useful inputs:

  • Story premise, news topic, campaign theme, or desired message.
  • Duration and format: 15s, 30s, 60s, 3min, vertical, horizontal, TV package, short video, or explainer.
  • Audience and platform.
  • Style: documentary, warm human story, fast social edit, policy explainer, youth culture, cinematic, official, or investigative.
  • Required facts, people, locations, slogans, products, or compliance constraints.

If duration is missing, assume 60-90 seconds for social video and state the assumption.

Workflow

  1. Restate the core story in one sentence.
  2. Choose a narrative structure:
    • Hook -> context -> character/problem -> turning point -> resolution -> call to action.
    • Problem -> evidence -> explanation -> solution -> takeaway.
    • Before -> change -> after.
  3. Divide the runtime into scenes or beats.
  4. For each beat, specify image, shot type, motion, narration/dialogue, sound, text overlay, and material needs.
  5. Add interview questions when human subjects are involved.
  6. Add AI-generation prompts only when the user wants assets for image/video tools.
  7. Check continuity: time, place, causality, character motivation, and tone.

Output Format

Use this structure by default:

## 创作定位
- 片名/暂定标题:
- 核心表达:
- 时长与比例:
- 风格:

## 分镜脚本
| 时间 | 镜号 | 画面/场景 | 景别/运动 | 旁白/对白 | 声音 | 字幕/包装 | 备注 |
|---|---:|---|---|---|---|---|---|

## 采访设计
- 采访对象:
- 关键问题:

## 素材清单
- 实拍:
- 资料:
- AI生成:
- 包装:

## 连贯性检查
- 逻辑:
- 人物:
- 节奏:
- 风险:

Quality Rules

  • Make every shot filmable or generatable.
  • Avoid vague instructions like "beautiful picture"; name the visible subject, action, environment, composition, and mood.
  • Keep narration speakable. Prefer short sentences.
  • Do not invent factual claims for news or public-policy content.
  • Flag missing source material when the script depends on unprovided facts, quotes, archive footage, or approvals.