# Shortdrama Asset Pack

> Generate a production-ready asset pack from a short-drama script, including Chinese image-generation prompts, script-named reference labels, character state variants, location GEO maps, prop variants, acting profiles, and voice locks. Use before storyboarding or video generation when the user needs reusable character, location, or prop assets rather than only an asset list.

- Skill: `towardsyoung/shortdrama-asset-pack` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add towardsyoung/shortdrama-asset-pack`
- Raw SKILL.md: https://api.skillmd.com/api/skills/towardsyoung/shortdrama-asset-pack/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: towardsyoung (https://skillmd.com/u/towardsyoung)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/towardsyoung/shortdrama-asset-pack

---


# Short-Drama Asset Pack

Turn a script into the reusable source of truth for later storyboard and video-prompt work. The output must contain prompts that can actually generate the assets, not only asset names or short descriptors.

This skill is standalone. It does not upload assets, call a generation API, or invent platform asset IDs.

## Inputs

Required:

- Script text, production outline, or a readable script file.

Optional:

- Story type and art style.
- Target image tools or platform.
- Target video model, such as Seedance 2.5.
- Existing asset images, platform `@tags`, or approved descriptions.
- Prompt language and output format.
- Whether to include acting or voice sections.

Defaults:

- All generation, edit, voice, acting, and video-prompt content is written in clear Chinese unless the user explicitly requests another language.
- Structural prompt headings such as `SCENE CONTEXT`, `ACTIVE REFERENCES`, `GEO SPATIAL LAYOUT`, `ACTION TIMING`, and `AUDIO` may remain in English.
- Include character, location, prop, state, acting, voice, and reference-manifest sections.
- Do not silently choose a visual style or model. Use the confirmation checkpoint below before producing assets.

## Mandatory Style and Model Checkpoint

Work in two phases.

### Phase 1: recommend and pause

When the user has not explicitly confirmed both the visual style and the model/tool route in the current request or immediately preceding context:

1. Read the script and identify its period, genre, tone, production difficulty, identity-continuity risk, material requirements, and likely delivery format.
2. Recommend one primary visual style and one primary model/tool route. The route may name separate image and video models when the workflow needs both.
3. Briefly explain why they fit this specific script and mention only material tradeoffs or platform-limit caveats.
4. Ask the user to confirm or revise the recommendation.
5. Stop. Do not generate the asset registry, asset prompts, storyboard, or video prompts in this phase.

Use a compact response:

```markdown
## 制作建议（待确认）
- 推荐视觉风格：...
- 推荐资产模型：...
- 推荐视频模型：...（仅在任务包含视频时）
- 推荐画幅与成片形态：...
- 推荐理由：...

请确认以上方案，或告诉我需要调整的风格、模型或画幅。确认后我再生成完整资产包。
```

### Phase 2: produce

Proceed only after the user confirms the recommendation or has already explicitly supplied both the visual style and model/tool route. Treat clear instructions such as “按推荐方案继续” as confirmation. Preserve the confirmed choices as the project source of truth and do not replace them later without asking.

## Label Contract

Keep these concepts separate:

- **Asset name**: the human-readable script name, such as `林小满` or `客厅`.
- **Suggested asset identifier**: a stable project-local handle based on the exact script name, such as `@林小满`. It is not proof that an asset has been uploaded.
- **Platform tag**: a real `@tag` supplied by the user or generation platform. Preserve it exactly; never rename or fabricate one.
- **Asset ID**: a platform or database identifier. Never generate one unless the user supplies it.

Suggested asset identifiers use the script's actual Chinese or original-language names, without generic type prefixes or transliteration:

- Character `何朝宗` becomes `@何朝宗`, not `@char_he_chaozong_base`.
- Location `何朝宗作坊` becomes `@何朝宗作坊`.
- Prop `观音瓷像` becomes `@观音瓷像`.
- Crowd `窑工` becomes `@窑工群体` when a group asset is required.

Append a short script-grounded state suffix only when needed, such as `@何朝宗_守窑`, `@观音瓷像_未烧紧握`, or `@龙窑_黎明开窑`. Use the shortest qualifier needed to avoid collisions. Never merge mutually exclusive states into one asset.

## Workflow

1. Complete the mandatory style and model checkpoint and record the confirmed choices.
2. Extract recurring characters, locations, plot-critical props, crowd groups, and every visible state change.
3. Separate stable identity from temporary state. Create one base asset plus independent state variants.
4. Define the confirmed project style as the source of truth. Keep character sheets neutral; put cinematic treatment in locations and video prompts.
5. Build the label registry from exact script names before writing prompts, reusing any real platform tags exactly.
6. Write all prompt content in Chinese by default and keep platform parameters separate. Never place aspect ratio, resolution, seed, or UI-only settings inside prompt prose.
7. For every recurring speaking character, write one acting master profile and one fixed voice prompt.
8. For every location, create one fixed GEO map and identify required reverse angles or state variants.

Read supporting references only when their output is needed:

- For character, location, prop, state-variant, and local-edit prompts, read [references/asset-prompts.md](references/asset-prompts.md).
- For recurring-character behavior and voice identity, read [references/acting-and-voice.md](references/acting-and-voice.md).

## Output Contract

Default Markdown output:

````markdown
已确认方案：{visual style, asset model, video model if relevant, aspect ratio}

## 项目资产生产包

### 全局视觉标准
| 项目 | 内容 |
|---|---|
| 故事类型 | ... |
| 成片风格 | ... |
| 角色图标准 | 中性灰背景、克制工作灯、真实皮肤、清晰眼神光 |
| 场景光照原则 | ... |
| 图像风格前缀 | ... |
| 视频风格前缀 | ... |

### 统一资产标识字典
| 资产标识或平台标签 | 标签状态 | 资产名称 | 类型 | 状态 | 首次出镜 | 使用范围 |
|---|---|---|---|---|---|---|

### 角色资产

#### {角色名}｜{标签}
**平台参数：** {model/UI settings; not prompt prose}
**身份锚点：** ...
**资产生成提示词：**
```text
...
```
**固定声音档案：** ...
**行为主档案：** ...
**状态变体：**
| 状态标签 | 唯一变化 | 生成/编辑方式 | 使用位置 |
|---|---|---|---|

### 场景资产

#### {场景名}｜{标签}
**平台参数：** ...
**资产生成提示词：** ...
**固定锚点：** ...
**GEO SPATIAL LAYOUT：** ...
**反向角度需求：** ...
**状态变体：** ...

### 道具资产
...

### 参考素材职责表
| 引用 | 标签状态 | 模态 | 对象 | 继承内容 | 不继承内容 | 使用范围 |
|---|---|---|---|---|---|---|
````

For long scripts, output the complete registry and all recurring assets once, then batch detailed prompts by episode or user-selected range. Do not silently omit assets.

## Asset Rules

- Use only story-relevant visual facts. Do not add decorative lore.
- A base character asset contains stable face, body, hair, marks, and baseline wardrobe. Injury, wetness, blood, aging, disguise, and costume changes are separate state assets.
- A location's day, night, rain, fire, destruction, or seasonal versions are separate assets when the visual state affects continuity.
- A prop's intact, open, broken, bloodied, concealed, or close-up version is separate when its silhouette or story function changes.
- Crowd members that do not need close-ups may be one crowd asset with explicit count range, height variation, and wardrobe variation. Important extras receive individual assets.
- Keep all references purpose-bound. State what each reference controls and what it must not control.
- Do not ask an image model to redraw a whole approved asset merely to add one local change. Use a minimal local edit and preserve everything else.
- Do not use real-person or protected-character names as visual shortcuts. Translate any user-provided inspiration into observable features unless the user is working with authorized material.

## Final Checks

- Every listed asset has a usable generation prompt or a clear reason no prompt is needed.
- UI parameters are separate from prompt prose.
- Suggested asset identifiers are visibly distinguished from real platform tags.
- Every temporary visual state is separated from the base asset.
- Every recurring speaking character has one fixed voice prompt.
- Every recurring character has one observable acting master profile.
- Every location has stable anchors, light direction, and a GEO map.
- Every reference has an explicit role and non-inheritance boundary.
- Every generated prompt is Chinese unless the user explicitly requested another language; permitted structural headings may remain English.
- Every suggested asset label uses the script's actual name and adds only a necessary state or collision qualifier.