film-character-asset

name	film-character-asset
description	Use when analyzing script characters for the current Banana/Veo short-drama pipeline. Output a dual-layer character asset spec: a slim control_table for downstream automation and a lightweight production_bible for human review and reuse.

Character Asset

Overview

This skill no longer writes long character analysis.

It must output a dual-layer structure:

• control_table
• production_bible

Rules:

• downstream automation reads only control_table
• production_bible is only for lightweight human review and reuse
• if the two layers conflict, control_table wins

Authority

This skill should follow the active project spec first.

Current mainline reference:

• ${OPENCLAW_HOME:-~/.openclaw}/agents/flashxiaolagu制作/production-optimization-spec.md

If the user is explicitly running another preserved baseline with its own local spec, that local spec can override this reference.

Core Principles

• infer only what is visually necessary and script-grounded
• keep only fields that directly stabilize asset generation and continuity
• remove personality prose, repeated explanation, and non-visual duplication
• keep the default asset state reusable, neutral, and easy to cut out
• production_bible must never introduce new hard facts that are missing from control_table
• keep character identity locks upstream, but keep board layout and aspect-ratio composition downstream in banana-character-prompt

When to Use

• at project start, after film-type-analysis
• when building stable character assets for image and video continuity
• when downstream prompts need hard visual control fields

Input

• full script text
• analysis/film-type-analysis.json

Output

• analysis/character-analysis.md
• analysis/character-analysis.json

Required Structure

Each character must contain:

• char_id
• name
• control_table
• production_bible

control_table keep only

• age_range
• face_structure
• hair
• costume_layers
• shoes_socks
• body_type
• stable_expression_baseline
• stable_asset_pose_baseline
• optional advanced locks only when they materially stabilize downstream asset generation:
  • portrait_eye_lock
  • beauty_calibration
  • height_reference_cm

production_bible keep only

• identity_label
• recognition_locks
• continuity_watchlist
• scene_state_notes

What To Remove

Do not keep these as default persisted output:

• long personality analysis
• long script-evidence blocks
• emotional interpretation
• repeated description of the same face / costume facts
• long continuity paragraphs

Writing Rules

1. face_structure must stay executable

Use compact visible fields only, such as:

• face shape
• eye shape
• iris color
• eyebrows
• nose
• mouth
• chin
• ears / ear accessories
• skin tone

2. hair must stay reusable

Keep only:

• length
• style
• parting / bangs
• color
• stable hair accessories

3. body_type must stay screen-readable

Keep only:

• height impression
• build
• shoulder line
• silhouette-relevant stable detail only if visible

4. costume_layers must stay layered

Keep only the default visible layers:

• upper layers
• lower layers
• stable accessories if they visually belong to the character baseline

5. stable baselines must be explicit

stable_expression_baseline must be a neutral reusable expression.

stable_asset_pose_baseline must default to:

• natural front-facing standing pose
• calm neutral expression
• both hands visible
• full body visible
• no story prop in hand unless it is part of the stable identity
• white / clean background for cutout reuse

6. production_bible must stay lightweight

Use only short bullets or short noun phrases.

Do not turn it back into a long production essay.

7. portrait_eye_lock is optional but useful

Use portrait_eye_lock only when portrait consistency matters or the role has a known risk of drifting into closed-eye beauty shots.

Good examples:

• 双眼清晰睁开，瞳孔完整可见
• 眼神平静清醒，不要闭眼，不要眯眼

This is a character-level visual lock, so it belongs upstream when needed.

8. beauty_calibration must stay restrained

Use beauty_calibration only when the role clearly needs elevated beauty control for downstream asset generation, such as:

• 知性优雅
• 高知感
• 高级美感
• 明星相
• 富养感 / 高门第感

Recommended compact values:

• natural
• refined
• refined_star_presence

Do not use it to invent a different face.

It only calibrates how polished and screen-ready the beauty level should be.

9. height_reference_cm is optional and must be grounded

Only add height_reference_cm when:

• the script explicitly gives a height
• the user gives a height
• downstream compositing requires a confirmed technical height reference

Do not guess a numeric height from vague prose.

If no confirmed numeric height exists, keep only body_type.height_impression.

10. Keep layout decisions downstream

Do not persist these in control_table:

• 9:16 or 16:9 board layout choice
• close-up panel position
• ruler placement
• reference-sheet composition wording

Those belong to banana-character-prompt, because they are generation-format decisions rather than stable character identity.

Output Format

Markdown

## character_001

**Control Table**
- age_range: 25岁左右
- face_structure:
  - 脸型：鹅蛋脸
  - 眼型：偏长杏眼
  - 瞳色：深棕
  - 眉毛：平直细眉
  - 鼻子：鼻梁细直，鼻尖小巧
  - 嘴：上薄下中等，嘴角自然下压
  - 下巴：圆润偏窄
  - 耳朵：耳垂小，细金耳钉
  - 肤色：自然白偏暖
- hair:
  - 长度：锁骨长度
  - 发型：顺直，三七侧分
  - 发色：深棕黑
- body_type:
  - 身高感：中等偏高
  - 体型：纤细但不单薄
  - 肩线：窄肩
- costume_layers:
  - 上装：米白针织开衫 + 浅杏圆领内搭
  - 下装：浅灰蓝高腰半裙
  - 稳定配饰：细金耳钉、细金项链、细表带腕表
- shoes_socks:
  - 裸色短袜
  - 浅米色低跟鞋
- stable_expression_baseline: 自然平静表情
- stable_asset_pose_baseline: 自然正面站姿，双手可见，全身完整，白底可抠图
- portrait_eye_lock: 双眼清晰睁开，瞳孔完整可见
- beauty_calibration: refined

**Production Bible**
- identity_label: 独立设计师 / 女主角
- recognition_locks:
  - 偏长杏眼
  - 锁骨长度深棕直发
  - 米白到浅杏色层叠穿搭
- continuity_watchlist:
  - 不要浓妆
  - 不要丢掉耳钉和腕表
  - 不要把针织外套改成西装
- scene_state_notes: 无

JSON

{
  "characters": [
    {
      "char_id": "character_001",
      "name": "林晓",
      "control_table": {
        "age_range": "25岁左右",
        "face_structure": {
          "face_shape": "鹅蛋脸",
          "eye_shape": "偏长杏眼",
          "iris_color": "深棕",
          "eyebrows": "平直细眉",
          "nose": "鼻梁细直，鼻尖小巧",
          "mouth": "上薄下中等，嘴角自然下压",
          "chin": "圆润偏窄",
          "ears": "耳垂小，细金耳钉",
          "skin_tone": "自然白偏暖"
        },
        "hair": {
          "length": "锁骨长度",
          "style": "顺直，三七侧分",
          "color": "深棕黑",
          "hair_accessories": []
        },
        "body_type": {
          "height_impression": "中等偏高",
          "build": "纤细但不单薄",
          "shoulder_line": "窄肩"
        },
        "costume_layers": {
          "upper_layers": [
            "米白色针织开衫",
            "浅杏色圆领内搭"
          ],
          "lower_layers": [
            "浅灰蓝高腰半裙"
          ],
          "stable_accessories": [
            "细金耳钉",
            "细金项链",
            "细表带腕表"
          ]
        },
        "shoes_socks": [
          "裸色短袜",
          "浅米色低跟鞋"
        ],
        "stable_expression_baseline": "自然平静表情",
        "stable_asset_pose_baseline": "自然正面站姿，双手可见，全身完整，白底可抠图",
        "portrait_eye_lock": "双眼清晰睁开，瞳孔完整可见",
        "beauty_calibration": "refined"
      },
      "production_bible": {
        "identity_label": "独立设计师 / 女主角",
        "recognition_locks": [
          "偏长杏眼",
          "锁骨长度深棕直发",
          "米白到浅杏色层叠穿搭"
        ],
        "continuity_watchlist": [
          "不要浓妆或夸张笑容",
          "不要丢掉耳钉和腕表",
          "不要把针织外套误生成为西装"
        ],
        "scene_state_notes": []
      }
    }
  ]
}

Constraints

1. Downstream automation reads only control_table.
2. production_bible must stay lightweight.
3. If the two layers conflict, control_table wins.
4. Output both markdown and json.
5. portrait_eye_lock / beauty_calibration / height_reference_cm are optional, not mandatory.
6. Layout and aspect-ratio composition instructions must stay downstream.