“image_input_policy”: {
“use_user_uploaded_image”: true,
“priority”: “OUTFIT_REFERENCE_ONLY”,
“freshness_scope”: {
“use_latest_upload_only”: true,
“ignore_any_prior_images_or_cached_context”: true,
“current_request_image_binding_required”: true
},
“binding”: {
“reference_image_required”: true,
“reference_image_role”: [
“outfit_source_only”,
“material_color_detail_reference”,
“accessory_reference”
],
“operation_type”: “GENERATE_NEW_MODEL_USING_OUTFIT_REFERENCE”,
“allowed_inferences”: [
“create_new_original_model_identity”,
“create_new_pose_and_body_proportions”,
“create_minimal_studio_background”,
“replicate_outfit_structure_from_reference”,
“replicate_color_palette_and_materials_from_reference”,
“replicate_accessories_if_visible”
],
“forbidden_reference”: [
“copy_face_identity_from_reference”,
“copy_pose_exactly_from_reference”,
“copy_body_proportions_exactly_from_reference”,
“infer_private_identity_details”
]
},
“fallback_if_no_image”: {
“mode”: “TEXT_ONLY_GENERATION”,
“subject”: “original_ai_modeled_young_chinese_influencer”,
“note”: “Only used when no user image is provided in the current request.”
}
},

“visual_director_thinking”: {
“spatial_logic”: “遵循‘从头到脚’的生物学穿戴顺序进行视觉排布。无论单品分布在模特左侧还是右侧，其垂直高度必须严格匹配穿戴顺序，确保标注箭头的逻辑指向自然且准确。”,
“labeling_protocol”: “尺码独占逻辑。全图中仅内衣单品允许出现‘Size’相关信息。其余所有单品仅标注单品名称，保持极简专业感。”,
“innerwear_set_enforcement”: “强制解构内衣套装（Bra + Panties），禁止仅展示上半身内衣。必须成套悬浮展示，并标注尺码。”,
“typography_variety”: “采用分级字体系统。主单品名称使用清晰中性无衬线体，内衣尺码使用更小、更具设计感的副标字体。配色方案根据穿搭智能调整，强调层次。”,
“deep_mastery_mode”: “必须进行深度思考：你同时是‘时尚大师 + 排版大师 + 设计师 + 平面设计师’。目标是：真实、克制、高级、可读、层级分明。优先保证已锁定硬约束不被破坏。”,
“consistency_audit_mode”: “生成后必须自检一致性：模特身上的每个可见单品（轮廓/颜色/材质/细节）与旁侧对应解构单品一致。若冲突，优先修正解构单品与标签，使其匹配模特；尽量不改动已正确的模特穿搭。”,
“diversity_engine”: {
“innerwear_diversity”: “内衣必须多样化：每次生成要在款式类别（bralette/underwire/sports/triangle/plunge/balconette）、材质（cotton/microfiber/satin/silk/lace/mesh）、细节（straps/back design/seamless/bow/minimal hardware）中显著变化，并与本次创建人物的性格气质匹配（例如：极简理性→无痕/运动；法式优雅→蕾丝/缎面；酷飒街头→运动/工装感）。避免连续多次出现同一类内衣视觉。”,
“bag_diversity”: “包包与内容物必须多样化：包型（tote/shoulder/crossbody/mini backpack/hobo）、材质（leather/nylon/canvas）、颜色与金属件风格需随人物气质变化。包内物品从不同类别组合中轮换（beauty/tech/stationery/lifestyle/snacks), 避免每次同一套物品。”,
“character_diversity”: “每次新建角色必须显著不同：至少在脸型轮廓、眉眼气质、发型长度/质感、妆容强弱、整体气场（冷感/甜感/中性/复古）上有明显差异；但单次画面内保持身份一致。”
}
},

“user_editable_fields”: [
{
“field_key”: “ootd_style_concept”,
“label_cn”: “穿搭核心风格 / 氛围描述”,
“description_cn”: “若有用户上传图片：图片为穿搭结构与配色来源；此字段用于补足命名与缺失细节（不得改变参考图的核心穿搭结构）。”,
“example_values”: [
“灰色极简主义瑜伽服，短背心，白色运动鞋”,
“黑色名媛风小香风套装，金属扣细节，搭配方头高跟鞋”,
“法式优雅：丝绸衬衫配灰色阔腿西装裤，搭配贝雷帽与珍珠项链”
]
},
{
“field_key”: “innerwear_with_size”,
“label_cn”: “内衣裤套装与尺码 (全图中唯一尺码标注项)”,
“description_cn”: “仍是全图唯一允许出现 Size 的标签。若未指定具体风格，AI 必须根据人物气质自行选择并确保多样性；若指定了具体内衣风格，则优先遵循但仍保证与人物匹配。”,
“example_values”: [
“Nude Seamless Silk Set | Size: 75B”,
“Black Minimalist Sporty Set | Size: M”,
“Premium White Lace Lingerie Set | Size: S”
]
}
],

“visual_logic_constraints”: {
“composition”: [
“STRICT MANDATORY 9:16 ASPECT RATIO”,
“NO BORDERS, NO MAIN TITLES, NO EXTERNAL FRAMES”,
“MANDATORY LARGE FULL BODY SHOT (80% height)”,
“MODEL MUST BE AI-CREATED ORIGINAL CHARACTER (DO NOT COPY PERSON FROM REFERENCE IMAGE)”,
“OUTFIT MUST FOLLOW REFERENCE IMAGE AS PRIMARY SOURCE”
],

“photorealism_hard_rules”: [
“PHOTOREALISTIC ONLY”,
“REALISTIC CAMERA LOOK: natural lens, realistic skin texture, real-world lighting”,
“NO cartoon, NO anime, NO manga, NO illustration, NO lineart, NO sketch, NO watercolor, NO 3D toon, NO vector icons-as-main-subject”,
“Text and UI elements must be crisp like magazine infographics, but the person and objects remain photographic”
],

“top_down_deconstruction_logic”: [
“Items MUST be arranged vertically based on wearing order: [Headwear/Accessories -> Outer Top -> Lingerie Set -> Bottoms -> Stockings/Socks -> Shoes].”,
“This order must be maintained even if items are split across left and right sides of the model.”,
“Mandatory Set: The lingerie set MUST include both a bra and matching panties.”,
“Exclusive Size Rule: NO size information allowed for any item except the Lingerie Set from {{innerwear_with_size}}.”
],

“deconstruction_coverage_rules”: [
“Aim for maximum coverage: each key worn item should have a corresponding separate deconstruction item shown.”,
“If the canvas becomes too crowded, merge only low-priority micro-accessories into a single grouped callout (e.g., ‘Jewelry set’) WITHOUT losing the main outfit pieces.”,
“Never omit Shoes, Top, Bottom, and the Lingerie Set callout. Headwear/accessories may be grouped only if necessary.”
],

“ui_elements”: [
“Upper-Right: Circular 2×2 Expression Chart under an arched serif ‘Expression Chart’ header.”,
“Expression Diversity: Each quadrant must show a noticeably different expression AND styling variation (hair/accessory changes). Keep identity consistent within this image.”,
“Sensory Zooms: Exactly 2 UNIQUE circular close-ups of BODY DETAILS (e.g., collarbone, glossy lips). No fabric zooms.”,
“Lifestyle Bag: Bottom-Right, LARGE bag with 5-7 items spilling out; exactly 3 major items labeled.”,
“Bag Spill Rule: Items must be visibly OUTSIDE the bag, scattered naturally with overlaps and contact shadows (no neatly placed grid, no floating-only).”,
“Typography: High-legibility horizontal sans-serif. Multi-color accents for hierarchy. Labels are clean and sharp.”
],

“annotations”: [
“Thin, elegant, non-overlapping black curved arrows.”,
“Arrows must point precisely to the correct location on the model’s body in a logical descending flow.”
],

“outfit_deconstruction_consistency_rules”: [
“For each visible worn item on the model, there must be a matching deconstruction item with the same silhouette, dominant color, and material impression.”,
“If mismatch occurs, FIX the deconstruction item visuals and labels to match the model first.”,
“Only if the model outfit deviates from the reference image’s core structure (wrong category), then adjust the model back toward the reference outfit.”
],

“text_qa_rules”: [
“STRICT TEXT WHITELIST: Only allow essential item name labels, the lingerie ‘Size’ text, and the exact header ‘Expression Chart’.”,
“NO meta or prompt-like text: forbid words/phrases such as ‘prompt’, ‘AI’, ‘generated’, ‘infographic template’, ‘layout’, ‘edit mode’, ‘reference image’, ‘do not’, ‘must’, or any instruction-like sentences.”,
“Language consistency: labels should match the overall design language; avoid random or out-of-context words.”
],

“character_and_item_diversity_rules”: [
“Innerwear must vary across generations: do not repeat the same innerwear type/material details too often.”,
“Bag and bag items must vary across generations: rotate bag silhouette/material and rotate item categories.”,
“Character face/hair/makeup vibe must vary across generations: avoid producing the same-looking model each time.”
],

“nanobanana_generation_stability”: [
“DO NOT output refusal or meta commentary. Proceed with photorealistic generation using outfit reference only.”,
“Avoid any claim about generating a specific real person; create an original character.”
],

“change_control”: [
“Do NOT modify already-correct sections: 9:16, no borders/titles, photorealism rules, bag spill scattered rule, arrow style, size-only-for-lingerie rule, lingerie set enforcement, expression chart placement, sensory zoom count.”,
“Only fix missing/incorrect parts: outfit vs deconstruction mismatches, accidental non-photorealistic drift, text whitelist violations, insufficient diversity in innerwear/bag/items/character.”
]
},

“final_image_prompt”: “MODE: OUTFIT-REFERENCE GENERATION (HIGHEST PRIORITY). Use ONLY the LATEST user-uploaded image in the CURRENT request as an outfit/material/accessory reference. Ignore any prior images or cached context. DO NOT copy the person from the reference image. Create an ORIGINAL AI model character through deep fashion/editorial thinking.\n\nROLE: You are simultaneously a fashion stylist, layout master, designer, and graphic designer. Make decisions like a top editorial art director: refined, balanced, readable, premium.\n\nOUTFIT REFERENCE SCOPE: Replicate the outfit structure, key accessories, color palette, and material cues from the reference image. Do not replicate the model identity, face, exact pose, or body proportions.\n\nPHOTOREALISM ONLY: Realistic photographic look with natural skin texture and real-world lighting. ABSOLUTELY NO cartoon/anime/manga/illustration/lineart/sketch/3D-toon styles. UI elements are crisp magazine-grade, but the person and objects remain photographic.\n\nCreate a professional 9:16 fashion editorial infographic with logical top-down deconstruction. [Subject]: An original, stylish young Chinese influencer (AI-created face) in a MANDATORY LARGE FULL BODY shot (80% height). [Outfit]: From the reference image; use {{ootd_style_concept}} only to name items or minimally infer unclear details WITHOUT changing the core outfit structure.\n\nINNERWEAR CHOICE (MANDATORY DIVERSITY): Generate exactly one matching lingerie set (Bra + Panties) from {{innerwear_with_size}}. If {{innerwear_with_size}} does not fully specify the style, choose a lingerie style that matches the created character’s vibe and ensure it differs clearly from common repeated defaults (vary silhouette/material/details). Size text appears ONLY for this lingerie set.\n\n[Background]: Unified seamless minimalist studio background, NO borders, NO main titles, NO external frames.\n\n[Layout Structure]:\n1) Top-Right: Circular 2×2 expression grid under an arched serif header ‘Expression Chart’. Each quadrant must show a noticeably different expression AND styling variation (hair/accessory changes), while keeping the same identity.\n2) Deconstruction Zones: Floating item callouts for Headwear/Accessories, Top, Lingerie Set, Bottom, Stockings/Socks (if present), Shoes. Items strictly arranged vertically by wearing order even if split left/right.\n3) Coverage Rule: Show each key worn item as its own deconstruction item when possible. If too crowded, group only micro-accessories into one callout; never omit Top/Bottom/Shoes/Lingerie.\n4) Sensory Zooms: Exactly 2 unique circular close-ups of BODY DETAILS (e.g., collarbone, glossy lips). No fabric zooms.\n5) Lifestyle Bag (DIVERSITY + SPILL): Bottom-right LARGE bag with 5-7 items SPILLING OUT. Choose a bag style/material/color that matches the character vibe and differs across generations. Items must be clearly outside the bag, scattered naturally with overlaps and contact shadows (no neat grid, no floating-only). Exactly 3 major items labeled. Bag items must vary across categories (beauty/tech/stationery/lifestyle/snacks) rather than repeating the same set every time.\n6) Annotations: Thin, elegant, non-overlapping black curved arrows, pointing precisely to correct locations in a logical descending flow.\n\nCONSISTENCY CHECK (MANDATORY): Ensure the model’s worn outfit and the side deconstruction items MATCH in silhouette, dominant color, and material. If mismatch occurs, correct the deconstruction items and labels to match the model first; do not change already-correct parts.\n\nTEXT QA (MANDATORY): Use ONLY essential item labels, the lingerie Size, and the header ‘Expression Chart’. Do NOT include any prompt-like or instruction-like text or out-of-context words.\n\n[Technical]: Photorealistic, ultra-sharp magazine-grade text, soft studio lighting, high resolution. Aspect Ratio: 9:16.”
}