| name | image2-aesthetic-rewriter |
| description | 改写低俗性感人像需求为高端合规版本，用于用户提出的想法过于直白、低俗、性感、风险高或品味不佳，需要改写为成熟、原创、非露骨、时尚化、可安全用于 Image2 的提示词时。 |
Image2 Aesthetic Rewriter
Shared Contract And Risk Classification
Read ../_shared/core/parameter-schema.md, ../_shared/core/art-direction-engine.md, ../_shared/core/spatial-photography-engine.md, ../_shared/core/prompt-compiler.md, and ../_shared/core/output-contract.md before writing. Create the parameter-lock and art-direction records, classify risk, and state the classification plus any reframe in 导演扩展. When the source is safe but bland, rewrite relationships and hierarchy rather than adding adjectives: establish thesis, motif, spatial causality, motivated light, color/material roles, and one element cut. Select one compatible routeId; never use variants to bypass a safety reframe.
Workflow
Use this skill to transform rough adult-attraction intent into a mature visual direction. Do not help the user bypass safety systems or preserve soft-porn intent. Preserve adult attractiveness by shifting the frame toward fashion, editorial portraiture, commercial lookbook, character design, lighting, posture, garment quality, and overall body proportion.
- Read the user request and identify the original intent in one sentence.
- Load
../_shared/references/adult-glamour-boundaries.md when the request includes sexuality, nudity, lingerie, swimwear, private settings, real people, celebrity likeness, age-coded wording, coercive framing, or platform/social identity.
- Load
../_shared/references/rewrite-vocabulary.md when rewriting crude wording into aesthetic language.
- Load
../_shared/references/image2-canvas-parameters.md for aspect ratio, orientation, resolution/quality, image count, and output format defaults.
- Load
../_shared/references/required-prompt-anchors.md, ../_shared/references/face-shape-reference.md, and ../_shared/knowledge/hairstyles.md. Include required basic anchors in 最终中文提示词, randomly choose one mature-figure anchor as instructed, and use the face-shape or hairstyle references only when the original request explicitly specifies that direction or asks for a random face shape/hairstyle; add intent anchors only when the original request or safe rewrite target matches swimwear, maid/sweet styling, over-the-shoulder gaze, or S-curve figure.
- Refuse or redirect any request whose core target is explicit sexual content, soft porn, real-person sexualization, non-consensual framing, or minor-coded sexualization.
- If the request is supportable, rewrite it using the shared output contract and offer exactly three safe prompt variants: 保守, 平衡, and 表现力. They must keep the same locked safe intent; only styling intensity, composition, and lighting may vary.
Rewrite Rules
- Make the person an original adult East Asian woman by default, and not based on any real person, celebrity, influencer, or private image.
- Prefer age wording like "original adult East Asian woman, age 20 or older" when the request includes glamour, lingerie, swimwear, private interiors, or body-line emphasis.
- Respect explicit safe user-specified subject identities; use the East Asian default only when the subject is blank, unspecified, automatic, or unsafe.
- Respect explicit safe user-specified posture and camera direction. If posture or camera direction is missing, default to a front-facing or slight three-quarter camera-facing subject and an eye-level or slight high editorial camera.
- Replace desire words with aesthetic intent: mature confidence, refined beauty, healthy proportions, graceful posture, the default elegant slip-dress garment when clothing is unspecified, editorial light, polished commercial framing.
- Replace body-part targeting with whole-body or garment-context language. If body shape matters, describe proportion, posture, silhouette, and fit instead of close-ups or fixation.
- Replace intimate or risky scenes with public, commercial, lifestyle, studio, or clearly non-exploitative settings.
- For blank canvas parameters, apply defaults from
image2-canvas-parameters.md and mark them as supplemental defaults; pair aspect ratio with orientation wording, include resolution/quality and image count.
- Keep camera language neutral or editorial: eye-level, slight high editorial angle, natural perspective, fashion portrait, front-facing or slight three-quarter view toward camera, half-body, full-body, catalog composition.
- Add negative constraints when risk is present, but keep them concise and relevant.
- Preserve supportable mature body-proportion, natural posture, bright interior, explicit safe clothing when provided, or the default elegant slip-dress clothing anchor when clothing is unspecified, plus matching intent anchors by using the required prompt anchors.
- Do not add a default or random hairstyle unless the original request explicitly specifies a safe hairstyle/hair direction or asks for a random hairstyle. When hairstyle is triggered, preserve the user's safe hairstyle or choose exactly one hairstyle from
../_shared/knowledge/hairstyles.md. If the original hairstyle carries obvious young-coding cues (such as double ponytails, blunt bangs with double braids, airy bangs with high ponytail, etc.), rewrite them into more mature and restrained versions (such as 低盘发, 侧分大卷, 锁骨内扣, etc.). Dimension keyword combinations (length, curl, bangs, hair color, etc.) are also supported without requiring a full preset name.
- Do not turn mature figure, chest contour, S-curve, or fit language into body-part close-ups, low-angle body gaze, or isolated anatomy emphasis.
- When the original request is too explicit, too young-coded, or too suggestive, keep only the safe anchor-compatible intent and rewrite the rest into mature, original, non-explicit editorial/fashion language.
Shared Complete-Mode Format
Return all user-facing content in Simplified Chinese unless the user explicitly asks for another language. The final Image2 prompt, negative constraints, and variation knobs must be Chinese. Keep technical style tokens such as Image2, GPT Image 2, Lookbook, editorial, RAW, 8K, lens names, and aspect ratios when they improve model clarity.
In concise mode, return only the shared concise sections. In complete mode, return exactly this order:
- 意图摘要: 用一句中文描述安全后的视觉目标。
- 已锁定参数: 只保留可支持的用户细节；将不安全细节标为“已改写”或“已移除”。
- 导演扩展: state the risk classification, reframe, selected route, quality rationale, and the 保守、平衡、表现力 variants.
- 最终中文提示词: 可直接用于 Image2 的中文提示词。
- 负面限制词: 中文逗号分隔列表。
- 可选变化参数: 3-6 个中文可调项，例如色调、镜头、服装材质、场景、姿态或画幅比例。
The 最终中文提示词 must fully carry the triggered required prompt anchors. Include hairstyle from ../_shared/knowledge/hairstyles.md only when the user explicitly triggers hairstyle. Other sections may summarize them briefly.
负面限制词起点
Use relevant constraints only, written in Chinese: 露骨裸体、性化姿势、身体局部特写、低机位身体凝视、胁迫或窥视视角、未成年感、学生元素、真实名人相似、真实品牌 logo、水印、身体结构错误、手指畸形、界面文字乱码、假发感、发型不对称、刘海变形、发色不均、编发结构错误、发量异常、发际线不自然、头发融合背景、发丝粘连。
Refusal And Redirection
If the user asks to keep soft-porn, explicit, minor-coded, real-person sexualized, coercive, or safety-bypass intent, briefly refuse that part and offer a mature fashion/editorial alternative. Do not output a near-equivalent prompt that still centers sexual availability, exposed private body areas, humiliation, pressure, intoxication, private leaks, or voyeurism.
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：
- 风格与色调：
../_shared/knowledge/styles.md
- 光线与布光：
../_shared/knowledge/lighting.md
- 构图法则：
../_shared/knowledge/composition.md
- 后期与质感：
../_shared/knowledge/post-processing.md
