image2-portrait-director

name	image2-portrait-director
description	调度 Image2 人像请求到对应视觉路线并生成合规中文结果，用于用户提出人像摄影需求但未明确具体风格或路线时，统一识别需求、选择最佳生成路径并输出合同级提示词。

Image2 Portrait Director

Required load order

Before making a decision, load these documents in order:

1. ../_shared/core/parameter-schema.md
2. ../_shared/core/reference-image-policy.md
3. ../_shared/core/conflict-resolution.md
4. ../_shared/core/art-direction-engine.md
5. ../_shared/core/spatial-photography-engine.md
6. ../_shared/core/prompt-compiler.md
7. ../_shared/core/quality-gates.md
8. ../_shared/core/output-contract.md
9. ../_shared/references/image2-canvas-parameters.md
10. ../_shared/knowledge/hairstyles.md
11. ../_shared/references/face-shape-hairstyle-guide.md
12. ../_shared/routes/route-registry.md
13. Exactly one matching file from ../_shared/routes/{routeId}.md.

Direction workflow

1. Parse user intent and lock safe explicit parameters according to the parameter schema.
2. Apply reference-image roles and conflict precedence before visual choices.
3. Build the art-direction record before route selection: one mild visual thesis, one motif, one decisive moment, viewer position, and three-level reading. Use the balanced-editorial default and preserve every lock.
4. Select 一个 Route (one primary Route) from the registry. Do not blend two main routes; compatible overlays may only add detail under the conflict-resolution contract.
5. Read the selected route, then convert its fingerprint, palette/material, camera, light, and photographed moment into a task-specific spatialPlan, color/material relationship, one restrained counter-expectation, and one element cut. Do not apply the route as a fixed prop bundle.
6. Delegate requests that explicitly ask for a fixed Chinese field list, structured portrait brief, or the field sequence "摄影风格 / 写真方向 / 场景方向 / 服装方向" to image2-structured-portrait-prompt. This specialist uses its own fixed output contract instead of the shared six-section contract.
7. Delegate other specialty work that the registry marks to image2-wedding-portrait, image2-fashion-lingerie-lookbook, image2-sports-venue-portrait, image2-character-poster-cover, image2-character-reference-sheet, or image2-risk-debugger. Their constraints outrank route styling.
8. Perform hairstyle routing based on user input and selected route:
   • If the user explicitly specifies a hairstyle, preserve the user's description; verify style compatibility with the selected route, and add a gentle note in the director expansion if mismatched.
   • If the user does not specify a hairstyle but mentions a face shape, recommend 1-2 hairstyles from ../_shared/references/face-shape-hairstyle-guide.md that best match both the face shape and the route style.
   • If the user specifies neither hairstyle nor face shape, recommend 1-2 hairstyles from the style route recommendation index in ../_shared/knowledge/hairstyles.md.
   • Hairstyle recommendations appear only as "造型建议" in the 导演扩展 section; do not add them to the final prompt by default unless the user says "加上" or the delegated specialist skill itself fills in hairstyles.
9. Apply spatial-photography causality, compile the final prompt by visual priority, apply quality gates, then return the selected skill's output contract in Simplified Chinese unless another language is requested.

导演扩展中的发型推荐

在完整输出模式的「导演扩展」部分，以「造型建议」子节的形式包含发型推荐，遵循以下规则：

• 推荐数量：1-2 款，每款含名称 + 一句话理由（例："法式慵懒卷——与法式优雅路线高度契合，八字刘海柔化脸部线条"）
• 用户已指定发型时：不做额外推荐，仅做风格匹配校验；如与选定路线不匹配，温和提示可考虑的替代方案，不强加修改
• 用户提到脸型时：追加「脸型适配说明」，解释推荐发型与脸型的适配逻辑（依据脸型搭配指南）
• 静默原则：发型推荐仅出现在「导演扩展」中，不默认写入「最终中文提示词」；用户明确要求"加上"时才纳入最终提示词

Image generation gate

Generate an image only when interactionMode: direct-image and the user explicitly says 直接出图. Otherwise return prompt text only. Reference images authorise only role-locked inputs; they do not authorise image generation.

参考知识库

生成提示词时可参考以下共享知识库模块（位于 _shared/knowledge/），根据用户需求选择合适的参数和表达方式：

• 人物特征：../_shared/knowledge/character-features.md
• 发型知识库：../_shared/knowledge/hairstyles.md
• 构图法则：../_shared/knowledge/composition.md
• 光线与布光：../_shared/knowledge/lighting.md
• 焦段与镜头语言：../_shared/knowledge/focal-length.md
• 姿势与体态：../_shared/knowledge/poses.md
• 场景与环境：../_shared/knowledge/scenes.md
• 服装与造型：../_shared/knowledge/fashion.md
• 风格与色调：../_shared/knowledge/styles.md
• 后期与质感：../_shared/knowledge/post-processing.md
• 脸型搭配指南：../_shared/references/face-shape-hairstyle-guide.md