playbooks

home / skills / zephyrwang6 / myskill / storyboard-generator

storyboard-generator skill

/storyboard-generator

This skill polishes your story, splits it into storyboard frames, selects a compatible visual style, and generates consistent, ready-to-use storyboards.

npx playbooks add skill zephyrwang6/myskill --skill storyboard-generator

Review the files below or copy the command above to add this skill to your agents.

Files (4)
SKILL.md
5.9 KB
---
name: storyboard-generator
description: 根据用户描述的故事内容，润色故事线并拆分为分镜脚本，批量生成风格一致的故事板插图。提供8种视觉风格选择，基于故事类型推荐最佳风格。调用Gemini API生成图片，保存到Obsidian图片目录。触发词："分镜故事"、"故事板"、"做个分镜"、"创建故事板"、"storyboard"、"画个故事"。
---

# 分镜故事板生成器

根据用户描述润色故事线，拆分为分镜脚本，批量生成风格一致的故事板插图。

## 工作流程

### Phase 1: 故事润色与分镜拆分

1. **接收故事** - 获取用户描述的故事内容（文字、大纲、或灵感片段）
2. **润色故事线** - 补充叙事细节，梳理情感弧线，输出结构化故事概要：
   ```
   故事概要：一句话概括
   情感弧线：[起点] → [转折] → [结局]
   ```
3. **拆分分镜** - 将故事拆为 6-10 个关键帧，每帧包含：
   - 场景标题
   - 镜头类型（远景/中景/近景/特写等）
   - 画面描述（人物动作、环境、光线）
   - 情绪基调
   - 对白/文字（如有）
4. **展示分镜脚本** - 以表格形式展示，等待用户确认或调整

展示格式：
```
## 分镜脚本

| # | 场景 | 镜头 | 画面描述 | 情绪 |
|---|------|------|---------|------|
| 1 | 开场 | 远景 | xxx | 平静 |
| 2 | 引入 | 中景 | xxx | 好奇 |
| ... |
```

### Phase 2: 风格选择

分镜脚本确认后，展示 8 种视觉风格（读取 references/visual-styles.md）：

| # | 风格 | 一句话特征 | 适用故事 |
|---|------|----------|---------|
| 1 | 赛博朋克漫画 | 霓虹+暗色调+高对比 | 科技/AI/未来 |
| 2 | 吉卜力水彩 | 柔和水彩+温暖光影 | 日常/温馨/成长 |
| 3 | 美式扁平插画 | 矢量几何+明快配色 | 产品/商业/教育 |
| 4 | 电影分镜手稿 | 黑白素描+镜头标注 | 视频脚本/广告 |
| 5 | 像素艺术 | 复古16-bit游戏风 | 游戏/趣味/极客 |
| 6 | 中国风水墨 | 水墨晕染+诗意留白 | 传统/哲理/东方 |
| 7 | 清线漫画 | 均匀粗线+扁平上色 | 冒险/幽默/IP |
| 8 | 写实概念艺术 | 电影级数字绘画 | 史诗/科幻/奇幻 |

根据故事类型和情感调性，**推荐 1-2 种最适合的风格并说明理由**，等待用户选择。

### Phase 3: 生成提示词

用户选定风格后，为每个分镜帧生成提示词。

每个提示词结构（读取 references/visual-styles.md 获取风格前缀和镜头语言）：
1. **风格前缀** - 选定风格的标准描述
2. **镜头指令** - 如 "Wide establishing shot" / "Close-up"
3. **画面叙述** - 用完整语句描述场景内容（人物、动作、环境、光线、氛围）
4. **情绪指令** - 明确画面传递的情感
5. **一致性锚定** - `IMPORTANT: This is panel [N] of [TOTAL] in a storyboard sequence. Maintain exact same art style, color palette, character design, and visual consistency with all other panels. Same character must look identical across all panels.`
6. **中文指令**（如需）- `ALL text, signs, and dialogue in the image MUST be in Chinese (Simplified Chinese). Technical terms may remain in English.`

展示提示词概要表，等待确认：
```
## 提示词概要

| # | 场景 | 镜头 | 风格 | 核心画面 |
|---|------|------|------|---------|
| 1 | 开场 | 远景 | 赛博朋克 | 城市天际线... |
| ... |

确认后开始生成（共 N 帧）
```

### Phase 4: 批量生成

确认后逐帧调用脚本生成：

```bash
python3 scripts/generate_image.py \
  --prompt "提示词" \
  --output "/Users/ugreen/Documents/obsidian/09image/MMDD-故事名-storyboard/01-场景名.png" \
  --api-key "AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8" \
  --api-url "https://generativelanguage.googleapis.com" \
  --model "gemini-3-pro-image-preview" \
  --aspect-ratio "16:9" \
  --resolution "2K"
```

**注意事项**:
- 故事板默认 **16:9 横版**
- 分辨率默认 **2K**
- MMDD 为当天日期
- 文件命名：`01-场景简称.png`、`02-场景简称.png`...
- 每帧间隔 2-3 秒避免限流
- 失败帧记录并继续，最后汇报
- **关键**: 角色外观描述在所有帧中保持一致（发型、服装、体型）

### Phase 5: 展示与迭代

1. 生成完成后，按顺序列出所有帧路径
2. 询问用户：
   - "整体故事节奏是否满意？"
   - "需要重新生成某一帧吗？"
   - "需要增加/删除某个场景吗？"
   - "角色一致性是否满意？"
3. 迭代：调整提示词重新生成指定帧

## 保存路径规范

```
/Users/ugreen/Documents/obsidian/09image/MMDD-故事名-storyboard/
├── 01-开场-远景.png
├── 02-引入-中景.png
├── 03-转折-近景.png
├── 04-发展A-中景.png
├── 05-发展B-全景.png
├── 06-高潮-特写.png
├── 07-结局-远景.png
└── storyboard-script.md  (可选：保存分镜脚本文档)
```

## API 配置

| 配置项 | 值 |
|-------|---|
| API URL | `https://generativelanguage.googleapis.com` |
| API Key | `AIzaSyDvvGGRbH4Os3Er0dYi0kE_AzE3_2b_Az8` |
| Model | `gemini-3-pro-image-preview` |
| 默认比例 | 16:9（故事板横版） |
| 默认分辨率 | 2K |

## 镜头语言速查

| 镜头 | 英文 | 用途 |
|------|------|------|
| 远景 | Wide Shot | 开场、环境交代 |
| 全景 | Full Shot | 人物全身+环境 |
| 中景 | Medium Shot | 对话、日常互动 |
| 近景 | Close-up | 表情、情感 |
| 特写 | Extreme Close-up | 关键道具、细节 |
| 俯拍 | Bird's Eye | 全局、渺小感 |
| 仰拍 | Low Angle | 权威、震撼 |

## 常见用法

| 用户说 | 操作 |
|-------|-----|
| "帮我画个故事分镜" | 收集故事 → 润色 → 拆分镜 → 选风格 → 生成 |
| "这个故事做成故事板" | 同上 |
| "用赛博朋克风画这个故事" | 跳过风格选择，直接用指定风格 |
| "第3帧不好，重新生成" | 调整提示词重新生成单帧 |
| "加一帧过渡镜头" | 在指定位置插入新分镜帧 |

Overview

This skill turns a user's story sketch into a polished storyboard sequence and batch-generates consistent storyboard illustrations. It refines the narrative, splits it into 6–10 storyboard panels with shot directions, and produces cohesive images in one of eight selectable visual styles. Generated images are saved to an Obsidian images folder and can be iterated on per-frame.

How this skill works

The skill first polishes the submitted story and extracts a concise story summary and emotional arc, then breaks the story into 6–10 panels with scene title, shot type, description, mood, and on-screen text. After confirming the panel script, it recommends 1–2 visual styles suited to the story, builds per-panel image prompts (style prefix, camera instruction, scene description, emotion, and an IMPORTANT consistency anchor), and invokes the Gemini image API to generate each frame. Images are saved to a dated Obsidian directory with consistent naming and resolution settings.

When to use it

  • You have a story outline and need a clear storyboard for production or pitching.
  • You want a quick set of cohesive concept images for a short film, animation, or comic.
  • You need style-consistent visuals for story pacing reviews or client presentations.
  • You want to iterate a single panel or re-generate failed frames without remaking the whole set.
  • You prefer automatic saving into an Obsidian image folder for easy project organization.

Best practices

  • Provide a clear prose description or bullet outline including main beats and key visual traits (character looks, locations, time of day).
  • Confirm or adjust the 6–10-panel split before generation to control pacing and content.
  • Select 1–2 recommended styles based on story tone; specify a preferred style to skip the recommendation step.
  • Include explicit character appearance details to ensure visual consistency across panels (hair, costume, body type).
  • Allow 2–3 second pauses between API calls to reduce rate-limit failures; review failure report and re-generate only failed frames.

Example use cases

  • Turning a short prose scene into a 7-panel storyboard for a director’s pitch.
  • Creating a 10-panel visual sequence in a Ghibli-watercolor style for a warm, coming-of-age short.
  • Generating a cyberpunk concept storyboard for a game pitch with neon-lit cityscapes.
  • Producing a black-and-white film-storyboard handoff for a commercial shoot.
  • Iteratively refining a single problem frame after client feedback while keeping other panels unchanged.

FAQ

Can I force a specific visual style?

Yes. If you specify a style (for example, Cyberpunk or Watercolor), the workflow skips recommendations and uses that style for all panels.

How are images named and where are they saved?

Images are saved to a dated Obsidian folder named MMDD-StoryName-storyboard with sequential filenames like 01-SceneShort.png, 02-SceneShort.png, etc., and a storyboard-script.md can also be saved.

What image settings are used by default?

Default outputs are 16:9 landscape, 2K resolution, and the Gemini image model; aspect ratio and resolution can be adjusted if needed.