Upload bench_eval_code/prompts/intelligent_prompts.json with huggingface_hub

bench_eval_code/prompts/intelligent_prompts.json

@@ -0,0 +1,77 @@
+{
+  "dimensions": {
+    "knowledge_reasoning": "You are an expert evaluator for knowledge-based reasoning image generation/editing tasks. In these tasks, the user provides a simple, goal-oriented instruction (and optionally a reference input image), and the AI model must invoke internalized domain knowledge to produce the output image.\n\nYou will be provided with:\n1. The user's **instruction**.\n2. The **domain** and **subtask** category of this task.\n3. The **output image** generated by the model.\n4. A **reference rationale** — an expert-written reasoning document illustrating ONE reasonable way to think through this task (the domain knowledge involved and the key information the result should convey).\n{ref_image_note}\n\nYour task is to evaluate the **Knowledge Reasoning Accuracy** of the output image: whether the model correctly invoked the relevant domain knowledge and produced content that is professionally accurate and comprehensive, fulfilling the reasoning intent behind the instruction.\n\n## How to use the reference rationale\n- The rationale is a **reference perspective, NOT an absolute standard**. It shows one expert's reasoning; it is not a mandatory spec to copy point-by-point.\n- Rely primarily on **your own domain expertise** and the domain-specific focus below. Use the rationale as a helpful hint for what correct reasoning and essential knowledge look like for this task.\n- If the output achieves the goal through a **different but equally valid** approach than the rationale describes, do NOT penalize it. Conversely, do NOT reward an image merely for superficially echoing the rationale's wording while lacking real domain correctness.\n- **Do NOT** judge visual quality, aesthetics, sharpness, typography legibility, or layout polish here — those are assessed separately. This dimension is ONLY about the correctness and completeness of the domain knowledge and reasoning intent.\n\n## Scoring Criteria (1-5):\n\n- **5 (Expert-level Reasoning)**: The output demonstrates deep domain expertise. All professional details are accurate and comprehensive. The content fulfills the reasoning intent and shows genuine understanding of the domain.\n  - Example: Instruction \"展示图中物件从二氧化硅到成品的完整加工流程\" → Correctly shows silica mining → melting → forming → annealing → cutting → polishing, with accurate process details.\n  - Example: Instruction \"给图中人物设计一个唐朝的妆容\" → Accurately presents Tang dynasty makeup features including 花钿 (forehead decoration), 斜红 (cheek marks), 面靥 (dimple decorations), with period-appropriate colors and styles.\n\n- **4 (Mostly Correct)**: The output invokes correct domain knowledge with the main content being accurate, but has minor omissions or imprecisions in details.\n  - Example: Manufacturing process is mostly correct but missing one intermediate step.\n\n- **3 (Partially Correct)**: The output shows some domain knowledge but contains noticeable errors or significant omissions.\n  - Example: Tang dynasty makeup includes some correct elements but mixes in obviously modern cosmetic styles.\n\n- **2 (Weak Knowledge)**: The output shows only surface-level understanding, lacking necessary professional knowledge support.\n  - Example: Manufacturing process only shows raw material and final product, with all intermediate steps missing.\n\n- **1 (Incorrect Knowledge)**: The output content severely contradicts domain knowledge, or completely fails to understand the instruction's intent.\n  - Example: Asked for Tang dynasty makeup but produces Western Gothic makeup.\n\n## Domain-Specific Evaluation Focus:\n\n{domain_focus}\n\n## Input\n**Domain:** {domain}\n**Subtask:** {subtask}\n**Instruction:** {instruct}\n\n**Reference Rationale (reference perspective — not an absolute standard):**\n{rationale}\n\n## Output Format\n\nProvide a detailed, step-by-step explanation of your evaluation, focusing on domain-knowledge correctness and reasoning intent (treating the rationale as a reference and explicitly ignoring visual/aesthetic quality). Conclude clearly with the final score wrapped in score tags, formatted as:\n\n<score>X</score>\n",
+    "visual_quality": "You are an expert image quality evaluator. You will be provided with an **output image** generated by an AI model for a knowledge-based reasoning task.\n\nYour task is to evaluate the **Visual Quality and Expressiveness** of the image, considering not just basic clarity but also the appropriateness of the visual format for the task.\n\n## Task Context\n**Domain:** {domain}\n**Subtask:** {subtask}\n**Instruction:** {instruct}\n\n## Scoring Criteria (1-5):\n\n- **5 (Professional-grade)**: The image is sharp, well-composed, and the visual style perfectly matches the task requirements. For multi-panel images (e.g., 4-grid, storyboards), the layout is clean with clear panel separation and logical flow. Text annotations (if any) are crisp and correctly placed. The overall visual effect is publication-ready.\n\n- **4 (Good Quality)**: The image quality is good with appropriate style. Multi-panel layouts are reasonable but individual panels may lack polish. Text annotations are mostly readable.\n\n- **3 (Acceptable Quality)**: The image has some quality issues (slight blur, distortion). Multi-panel formats may not be well-organized or panels are hard to distinguish. Text annotations are partially unreadable.\n\n- **2 (Poor Quality)**: The image is noticeably blurry or distorted. Multi-panel layouts are chaotic. Text is severely deformed and unreadable.\n\n- **1 (Very Poor Quality)**: The image is severely distorted, unrecognizable, or completely fails to match the required visual format.\n\n## Special Evaluation Points:\n- **Multi-panel tasks** (comics, storyboards, process flows, travel photos): Is the panel count correct? Are panels clearly separated? Is the narrative flow coherent across panels?\n- **Annotation tasks** (structural labeling, nutrition analysis, landmark identification): Is the text legible? Are labels correctly positioned?\n- **Chart/diagram tasks** (data visualization, relationship graphs): Is the chart type correct? Is the data presentation clear?\n- **Design tasks** (sketches, posters, business cards): Does it follow professional design conventions and aesthetics?\n\n## Output Format\n\nProvide a detailed evaluation. Conclude clearly with the final score wrapped in score tags, formatted as:\n\n<score>X</score>\n",
+    "input_consistency": "You are a precise image consistency evaluator. You will receive:\n\n1. **Original Image(s)**: One or more reference input images provided by the user.\n2. **Output Image**: The image generated/edited by the AI model.\n3. **Instruction**: The user's modification/generation instruction.\n\nYour task is to evaluate whether the output image maintains consistency with the input image(s) in aspects that should be preserved, while correctly applying the instructed changes.\n\n## Task Context\n**Domain:** {domain}\n**Subtask:** {subtask}\n**Instruction:** {instruct}\n\n## Scoring Criteria (1-5):\n\n- **5 (Perfect Consistency)**: All elements that should be preserved from the input image(s) are perfectly maintained. The core subject (person's appearance, object's shape, room layout, etc.) is identical to the input.\n\n- **4 (Minor Deviation)**: Core elements are well preserved with only minor, negligible differences in secondary details.\n\n- **3 (Noticeable Deviation)**: Core elements are partially preserved, but there are obvious inconsistencies that a careful observer would notice.\n\n- **2 (Significant Deviation)**: Core elements have substantially changed. The output barely resembles the input in key aspects.\n\n- **1 (No Consistency)**: The output has almost no visual connection to the input image(s). Core subjects are completely different.\n\n## Domain-Specific Consistency Focus:\n- **Makeup/Hairstyle tasks**: The person's facial features (face shape, skin tone, eye shape) must remain consistent.\n- **Home renovation/Storage**: The room structure, dimensions, and fixed elements (walls, windows) must be preserved.\n- **Product processes/Disassembly**: The original product/object must be correctly identified and maintained.\n- **Historical comparison**: The landmark/location must be correctly recognized.\n- **Game character tasks**: The character's visual identity must be preserved.\n- **Perspective transformation**: Core scene elements must be consistent across viewpoint changes.\n- **Travel photos with person**: The person's appearance must be consistent across different scenes.\n\n> Note: Changes explicitly required by the instruction should NOT be counted as inconsistencies.\n\n## Output Format\n\nProvide a detailed evaluation. Conclude clearly with the final score wrapped in score tags, formatted as:\n\n<score>X</score>\n"
+  },
+  "generic_focus": "Evaluate based on general domain knowledge accuracy, professional correctness, and completeness relevant to this task.",
+  "task_focus": {
+    "地理与历史-历史人物社交": "### 历史人物社交 (Historical Figure Social Media)\n- Does the social media profile **accurately reflect the historical figure's identity** (correct portrait, era-appropriate description)?\n- Are the **platform-specific UI elements** correctly reproduced (LinkedIn layout, Instagram grid, Keep workout interface)?\n- Is the **content contextually humorous/creative** while remaining historically grounded?\n- Are **biographical details** (titles, achievements, era) factually accurate?\n- Does the design show understanding of **both the historical figure and the modern platform**?",
+    "地理与历史-历史路线地图": "### 历史路线地图 (Historical Route Map)\n- Is the **route geographically accurate** (following known historical paths, not impossible shortcuts)?\n- Are **key waypoints and destinations** correctly marked (e.g., 玄奘取经 should pass through Central Asia, India)?\n- Is the **map style** appropriate (historical cartography aesthetic, not modern Google Maps)?\n- Are **geographic features** (mountains, rivers, seas) correctly positioned along the route?\n- For military campaigns: Are **battle sites and conquered territories** accurately marked?",
+    "地理与历史-地标识别标注": "### 地标识别标注 (Landmark Identification & Annotation)\n- Are **all major landmarks** in the image correctly identified by name?\n- Are the **names accurate** (official names, not colloquial or incorrect)?\n- Are annotation labels **clearly positioned** with lines/arrows pointing to the correct buildings?\n- Is the **text legible** and not overlapping with other annotations?\n- For functional zone labeling (e.g., 古罗马斗兽场功能分区): Are the **zones historically accurate** (arena floor, seating tiers, underground passages, emperor's box)?",
+    "地理与历史-坐标场景还原": "### 坐标场景还原 (Coordinate-based Scene Reconstruction)\n- Is the **geographic location correctly decoded** from the coordinates (e.g., 48.85°N, 2.29°E = Eiffel Tower area, Paris)?\n- Does the scene show **recognizable features** of that specific location?\n- Is the **atmosphere/mood** as specified (e.g., \"法式浪漫\" = romantic Parisian scene, \"人潮涌动\" = crowded bustling scene)?\n- Are **local architectural styles, vegetation, and cultural elements** accurately depicted?\n- For historical event locations: Is the **historical context** correctly represented?",
+    "地理与历史-时代对比": "### 时代对比 (Era Comparison)\n- Is the **historical period accurately identified** for the location shown (e.g., Shanghai Bund in the 1920s vs. modern)?\n- Are **period-specific architectural details** correct (building styles, materials, heights)?\n- Are **transportation methods** era-appropriate (horse carriages, rickshaws, early automobiles for early 1900s)?\n- Are **clothing styles of pedestrians** historically accurate?\n- Is the **same geographic viewpoint** maintained between historical and modern versions?\n- For specific era requests (e.g., \"乾隆盛世\", \"福尔摩斯的年代\"): Are the **cultural and temporal details** precisely matched to that specific period?",
+    "影视与叙事-小说段落": "### 小说段落 (Novel Scene Visualization)\n- Are the **specific scene details** from the novel accurately depicted (e.g., 黛玉葬花 should show Lin Daiyu burying fallen petals with a flower hoe)?\n- Are **character appearances** consistent with their canonical descriptions in the novel?\n- Is the **setting/environment** accurate to the novel's world (e.g., Grand View Garden for 红楼梦, Heavenly Palace for 西游记)?\n- Does the image capture the **emotional tone** of the specific scene (tragic, heroic, comedic)?\n- Are **period-appropriate costumes, architecture, and props** depicted?",
+    "影视与叙事-漫画叙事": "### 漫画叙事 (Comic Narrative)\n- Does the comic tell a **complete story** with beginning, middle, and end within the specified panel count (e.g., 4格漫画)?\n- Is the **narrative logic** clear without requiring external explanation?\n- Are **character designs consistent** across all panels (same clothing, features, proportions)?\n- Does the comic use proper **comic language** (speech bubbles, sound effects, motion lines, panel gutters)?\n- For character-based comics: Does the protagonist **match the reference image** in appearance?\n- For historical/cultural stories (e.g., \"亡灵节的由来\", \"三顾茅庐\"): Are the **historical/cultural facts accurate**?\n- Is there appropriate **humor, drama, or educational value** as implied by the instruction?",
+    "影视与叙事-电影分镜": "### 电影分镜 (Movie Storyboard)\n- Does each panel represent a **distinct shot** with clear camera angle indication (close-up, wide shot, over-the-shoulder, bird's eye)?\n- Is the **narrative arc** coherent (setup → conflict → climax → resolution) within the specified number of panels?\n- Are **shot transitions** logical (e.g., establishing shot before close-up, reaction shots after action)?\n- Do the panels follow **professional storyboard conventions** (panel numbering, action arrows, camera movement notes)?\n- Are character positions, expressions, and actions **consistent across panels** (continuity)?\n- Does the visual storytelling convey the **theme/mood** specified in the instruction (e.g., heroic, suspenseful, comedic)?",
+    "影视与叙事-进化演变": "### 进化演变 (Evolution/Progression)\n- Are **all major evolutionary stages** represented with historically/scientifically accurate examples?\n- Is the **chronological order strictly correct** (e.g., for phones: brick phone → flip phone → early smartphone → modern smartphone)?\n- Does each stage show **visually distinct characteristics** that clearly differentiate it from adjacent stages?\n- Are the **time periods or era labels** accurate (e.g., \"1980s brick phone\" should look like a Motorola DynaTAC, not a Nokia 3310)?\n- For biological evolution: Are the morphological changes scientifically plausible?\n- For technological evolution: Are key innovations (e.g., touchscreen, color display) shown at the correct historical point?",
+    "数据与信息-��系图谱": "### 关系图谱 (Relationship Graph)\n- Are **all major characters/entities** included in the graph?\n- Are the **relationships correctly identified** (family ties, alliances, rivalries, hierarchies)?\n- Is the **relationship type** clearly labeled on connecting lines (师徒, 父子, 对手, 盟友)?\n- For classification tasks: Are entities **correctly categorized** into the specified groups (e.g., 凡人科技/肉体变异/天神魔法 for Marvel characters)?\n- Is the **graph layout** clear and readable (not a tangled mess of crossing lines)?\n- Are **character portraits or icons** included for visual identification?\n- For literary works: Are relationships **consistent with the source material** (e.g., 红楼梦 family tree should match the novel)?",
+    "数据与信息-数据可视化": "### 数据可视化 (Data Visualization)\n- Is the **chart type appropriate** for the data (pie chart for proportions, line chart for trends, bar chart for comparisons, radar chart for multi-dimensional comparison)?\n- Are **data values accurate** and consistent with the source information?\n- Are **axes properly labeled** with units and scale?\n- Is the **color scheme** effective for distinguishing data categories?\n- For derived visualizations (e.g., \"全场最佳球星卡\"): Does the design **creatively integrate** the data into an engaging visual format?\n- Are **legends, titles, and annotations** present and correctly placed?\n- For trend data: Is the **time axis** correctly scaled and labeled?\n- Does the visualization tell a **clear data story** that matches the instruction's intent?",
+    "数据与信息-文档流程化": "### 文档流程化 (Document-to-Flowchart)\n- Are **all major process steps** extracted from the source document?\n- Is the **step sequence logically correct** (no steps out of order or missing critical dependencies)?\n- Are **decision points and branches** correctly identified (if the process has conditional paths)?\n- Does the flowchart use **standard flowchart symbols** (rectangles for processes, diamonds for decisions, arrows for flow)?\n- Is the **level of detail** appropriate (not too granular or too abstract)?\n- For complex processes (e.g., semiconductor manufacturing, vaccine production): Are **technical terms** correctly used?\n- Are **parallel processes** correctly represented (if applicable)?\n- Is the flowchart **readable** with clear flow direction and no crossing arrows?",
+    "数据与信息-信息图": "### 人物信息图 (Biographical Infographic)\n- Are the **biographical facts** historically accurate (dates, achievements, titles, affiliations)?\n- Is the **timeline chronologically correct** with major milestones properly ordered?\n- Does the infographic use **appropriate visual metaphors** for the person's field (scientific diagrams for scientists, product timelines for entrepreneurs, military maps for emperors)?\n- Is the **visual style** consistent with the person's era and identity?\n- Are **data points and statistics** (if any) factually verifiable?\n- Does the layout follow **professional infographic design principles** (visual hierarchy, consistent iconography, readable typography)?",
+    "时尚与美妆-妆容改进": "### 妆容改进 (Makeup Improvement)\n- Are the **specific issues** in the original makeup correctly identified and addressed?\n- For \"看着更漂亮/更年轻\": Are the makeup adjustments **age-appropriate and enhancing** (e.g., brightening under-eye area, adding subtle blush, defining brows)?\n- Are the makeup changes **realistic and achievable** with standard cosmetics (not digitally altered skin)?\n- Does the improved makeup maintain the person's **natural features** while enhancing them?\n- For annotation tasks: Are the **problem areas correctly identified** with clear visual markers?",
+    "时尚与美妆-妆容迁移": "### 妆容迁移 (Makeup Transfer)\n- Does the makeup accurately reflect the **specified era/culture/occasion**?\n  - **唐朝妆容**: Should include 花钿 (forehead floral decoration), 斜红 (diagonal cheek marks), 面靥 (dimple dots), thick eyebrows (蛾眉), red lip color, white face powder\n  - **狂欢节妆容**: Should feature bold colors, glitter, feathered elements, theatrical eye makeup, festive patterns\n  - **舞会妆容**: Should be elegant and glamorous with smoky eyes, defined contours, statement lip color\n  - **明星妆容**: Should be polished, camera-ready with flawless base, sculpted features, dramatic eyes or lips\n- Is the person's **facial structure and skin tone preserved** while only the makeup changes?\n- Are the makeup elements **historically/culturally authentic** (not a modern interpretation unless specified)?",
+    "生活与消费-家居生活": "### 家居生活 (Home Living Scene)\n- Does the scene realistically depict the **home-living activity/scenario** described?\n- Is the **room structure and key fixed elements preserved** while the intended changes are applied?\n- Are objects placed **plausibly and functionally** within the space?\n- Is the **atmosphere/style** consistent with the instruction (温馨/整洁/某风格)?\n- Does the result look like a real, livable space?",
+    "生活与消费-旅游出行": "### 旅游出行 (Travel/Tourism)\n- Are the **landmark locations accurate** for the specified destination (e.g., Eiffel Tower for Paris, Great Wall for Beijing)?\n- Do the scenes show **iconic and recognizable** tourist spots, not generic cityscapes?\n- For person-based travel photos: Is the **person's appearance consistent** across all sub-images?\n- Does each sub-image show a **distinct location/activity** (not repetitive scenes)?\n- Is the **visual style** appropriate for the platform (e.g., \"小红书打卡\" should have a lifestyle/aesthetic quality)?\n- Are **cultural elements** of the destination authentically represented (local architecture, food, customs)?\n- Is the specified number of sub-images correct (e.g., \"4个子图\" = exactly 4 images)?",
+    "生活与消费-美食制作": "### 美食制作 (Cooking Process)\n- Are the cooking steps in **correct culinary order** (e.g., prep ingredients → marinate → heat oil → stir-fry → season → plate)?\n- Does each step show a **visually distinct stage** of the cooking process?\n- Are the **ingredients recognizable** and consistent with the dish being prepared?\n- Are **cooking techniques** accurately depicted (e.g., julienne cutting, blanching, sautéing vs. deep-frying)?\n- Does the final result look like the **actual dish** shown in the reference image?\n- Is the specified number of steps/panels correct?",
+    "生活与消费-导购对比": "### 导购对比 (Product Comparison)\n- Does the comparison chart/infographic extract the **core differentiating parameters** (e.g., price, specs, materials, target audience) from the product descriptions?\n- Are comparison dimensions **fair and balanced**, not biased toward one product?\n- Is the visual layout organized as a structured comparison (e.g., side-by-side table, radar chart, feature matrix) rather than random text?\n- Are product images correctly associated with their respective descriptions?\n- Does the comparison highlight **key trade-offs** that help a consumer make an informed decision?",
+    "科学与教育-数学题解答": "### 数学题解答 (Math Problem Solving)\n- Is the **solution process logically correct** with proper mathematical reasoning?\n- Are **all intermediate steps** shown (not jumping from problem to answer)?\n- Is the **final answer numerically correct**?\n- Are **mathematical notations** properly used (symbols, equations, geometric constructions)?\n- For geometry problems: Are **auxiliary lines, angle marks, and labels** correctly drawn?\n- For pattern/logic problems: Is the **reasoning pattern** correctly identified and explained?\n- Does the visual presentation make the **solution process easy to follow**?",
+    "科学与教育-物理现象推理": "### 物理现象推理 (Physics Reasoning)\n- Are **all relevant forces** correctly identified and represented (gravity, normal force, friction, tension, applied force)?\n- Are **force vectors** drawn with correct direction, relative magnitude, and point of application?\n- Is the **physical scenario** correctly interpreted (e.g., inclined plane, pulley system, projectile motion)?\n- Are **Newton's laws** correctly applied in the force analysis?\n- For dynamic scenarios: Are **acceleration direction and motion trajectory** correctly indicated?\n- Are **standard physics diagram conventions** followed (labeled arrows, coordinate axes, free body diagrams)?",
+    "科学与教育-科普图解": "### 科普图解 (Science Illustration)\n- Are the **scientific principles** accurately depicted (e.g., photosynthesis should show light absorption, CO₂ intake, O₂ release, glucose production)?\n- Is the **level of detail** appropriate for educational purposes (simplified but not incorrect)?\n- For cellular/molecular processes: Are **organelles and molecules** correctly identified and positioned?\n- For astronomical phenomena: Are **scale relationships and physical mechanisms** correctly shown (e.g., tidal forces from moon's gravity)?\n- Is the **visual style** engaging and educational (clear labels, color coding, directional arrows)?\n- For personified/cartoon-style illustrations: Do the characters **accurately represent** the scientific entities they portray?",
+    "科学与教育-结构标注": "### 结构标注 (Structural Annotation)\n- Are **all major components** correctly identified with proper technical terminology?\n- Are annotation lines/arrows **clearly pointing** to the correct parts (not ambiguous)?\n- Is the **naming convention** consistent with the field (e.g., 弦轴 for violin tuning pegs, 共鸣箱 for resonance chamber)?\n- Are components labeled in a **logical grouping** (e.g., by system/function)?\n- Is the **text legible** and not overlapping with the image or other labels?\n- For biological structures: Are **all organelles/parts** at the correct relative positions and sizes?",
+    "设计与产品-使用流程": "### 使用流程 (Usage Flow)\n- Does the flow cover the **complete user journey** from setup/preparation to final use?\n- Are the steps in **correct operational order** (e.g., for a coffee machine: fill water → add beans → select mode → brew → serve)?\n- Does each step show a **distinct, meaningful action** rather than trivial sub-steps?\n- Are safety precautions or important tips included where relevant?\n- Is the number of steps/panels as specified in the instruction (e.g., \"4宫格图\" = exactly 4 panels)?",
+    "设计与产品-加工流程可视化": "### 加工流程可视化 (Manufacturing Process Visualization)\n- Are **all critical manufacturing stages** present (e.g., raw material → processing → forming → finishing → quality control → final product)?\n- Is the **sequence strictly correct** according to real-world industrial processes?\n- Does each stage show the **correct transformation** of the material (e.g., silica melting into glass, metal being forged)?\n- Are specialized tools, equipment, or techniques (e.g., annealing furnace, CNC machining) accurately depicted?\n- Is the visual flow direction clear (left-to-right or top-to-bottom) with logical transitions between stages?",
+    "设计与产品-名片设计": "### 名片设计 (Business Card Design)\n- Does the card include **contextually appropriate information** for the person (e.g., historical figure's title, achievements, era)?\n- Is the **visual style** consistent with the person's identity (e.g., classical Chinese style for Cao Cao, scientific motifs for Einstein)?\n- Does the layout follow **professional business card conventions** (name prominence, contact info hierarchy, balanced whitespace)?\n- Are typography choices appropriate (font style, size hierarchy, readability)?\n- Is the card design creative yet professional, avoiding cluttered or amateurish layouts?",
+    "设计与产品-广告海报设计": "### 广告海报设计 (Advertising Poster Design)\n- Does the poster effectively communicate the **product/brand message** with a clear visual hierarchy?\n- Is the **target audience** appropriately addressed through visual style, tone, and imagery?\n- Does the composition follow **professional advertising conventions** (headline, visual focal point, call-to-action, brand logo placement)?\n- Are **color psychology and typography** used effectively to evoke the desired emotional response?\n- Does the poster balance **aesthetic appeal with commercial effectiveness**?",
+    "设计与产品-结构拆解": "### 结构拆解 (Structural Disassembly)\n- Are **all major components and sub-assemblies** correctly identified and separated?\n- Is the **spatial relationship** between parts preserved (i.e., an exploded view should show where each part connects)?\n- Are parts labeled with **correct technical names** (e.g., \"crankshaft\" not just \"metal rod\")?\n- Does the disassembly follow a **logical order** (outer shell → internal mechanisms → core components)?\n- For vehicles/machines: Are key systems (drivetrain, suspension, electrical) correctly represented?",
+    "设计与产品-设计流程": "### 设计流程 (Design Process)\n- Does the flow show the **complete design lifecycle** (concept sketch → refinement → prototyping → final product)?\n- Is each stage visually distinct and showing **progressive refinement** of the design?\n- Are industry-standard design phases represented (ideation, CAD modeling, material selection, manufacturing)?\n- Does the progression demonstrate realistic **evolution from abstract to concrete**?\n- Is the final product stage polished and production-ready in appearance?",
+    "设计与产品-设计草图": "### 设计草图 (Design Sketch)\n- Does the sketch follow **professional industrial design conventions** (e.g., orthographic views, perspective drawings, exploded views)?\n- Are **proportions and scale** accurate relative to the real object?\n- Does the sketch include appropriate **annotation lines, dimension markings, or material callouts**?\n- Is the drawing style consistent with professional design drafting (clean lines, proper shading, cross-hatching for materials)?\n- Does the sketch capture the **essential functional and aesthetic features** of the object?",
+    "IP与娱乐-周边套装设计": "### 周边套装设计 (Merchandise Set Design)\n- Are the merchandise items **functionally diverse** (not just the same design on different objects)?\n- Do all items share a **cohesive visual theme** derived from the IP?\n- Are the **IP's core visual elements** abstracted appropriately for merchandise (not just slapping a character image on products)?\n- Are the product types **practical and marketable** (stationery, accessories, apparel, collectibles)?\n- Does the design quality suggest **official licensed merchandise** rather than fan-made items?",
+    "IP与娱乐-主题空间设计": "### 主题空间设计 (Themed Space Design)\n- Does the design **authentically capture the specified theme/IP** (e.g., 《黑客帝国》= green digital rain, dark industrial aesthetic, pod-like furniture; 蝙蝠侠基地 = cave-like space, dark tones, tech equipment, bat motifs)?\n- Are **theme-specific furniture, lighting, and decorative elements** present?\n- Is the space **functionally usable** as the specified type (避难所 should have survival equipment, 基地 should have command center elements)?\n- Does the design maintain **architectural plausibility** (walls, ceiling, floor are structurally sound)?\n- Is the **atmosphere and mood** consistent with the theme (dark and techy for sci-fi, serene for 日式禅意)?\n- For renovation tasks: Is the **original room structure preserved** while the theme is applied?",
+    "地理与历史-卫星图场景化": "### 卫星图场景化 (Satellite-to-Scene Conversion)\n- Are the **geographic features** from the satellite image correctly interpreted (buildings, roads, water bodies, vegetation)?\n- Does the ground-level scene **match the layout** visible in the satellite view?\n- For style transfers (赛博朋克, RPG风格): Are **style-specific visual elements** applied while maintaining geographic accuracy?\n- Is the **scale and density** of buildings/features consistent with the satellite image?\n- Are **lighting conditions** appropriate for the specified time/mood (e.g., \"深夜场景\" = nighttime lighting)?",
+    "生活与消费-户型改造与空间重构": "### 平面图立体化 (Floor Plan to 3D)\n- Does the 3D rendering **accurately reflect the 2D plan's room layout** (room sizes, wall positions, door/window locations)?\n- Are **furniture and fixtures** placed in positions consistent with the floor plan?\n- Is the **interior design style** appropriate for the building type (精装房 = modern luxury, 院落 = traditional Chinese courtyard)?\n- Are **proportions and scale** correct (ceiling height, room dimensions)?\n- For historical buildings: Does the 3D reconstruction reflect the **building's era-appropriate architectural style**?\n- Is the **specified viewpoint** correct (e.g., \"从露台视角\" = view from the terrace)?",
+    "IP与娱乐-IP资料": "- Does the generated document (resume / profile / IP dossier) present the character's identity in a **role-appropriate, professionally formatted** way (e.g., a film-audition resume lists relevant roles, skills, physical stats; a LinkedIn-style page uses correct section layout)?\n- Are the **fabricated but plausible details** consistent with the character's visual identity and implied background (era, profession, temperament)?\n- Is information organized with clear hierarchy (name/title header, sections, bullet points) rather than a wall of text?\n- Does the design match the intended platform/genre conventions (formal resume vs. social profile vs. character wiki card)?\n- Is the character's likeness from the reference image preserved in any portrait/avatar used.",
+    "IP与娱乐-卡牌游戏设计": "- Do the **stat distributions and class identity match the character archetype** (e.g., a melee mage emphasizes INT + close-range traits; an agility assassin emphasizes AGI/crit, low HP)?\n- Are **standard game-panel elements** present and correctly laid out (portrait, name, class/role, HP/MP/ATK/DEF/AGI or game-specific stats, skill/ability area, rarity indicator)?\n- If a specific game IP is named (e.g., 魔兽世界, 暗黑破坏神), does the **UI style, terminology and stat framework** match that game's conventions?\n- Are stat values numerically balanced and internally consistent (not all maxed)?\n- Is the character portrait consistent with the input image and dynamic/striking enough for a collectible card.",
+    "IP与娱乐-宣发材料": "- Does the promotional material (movie poster / marketing key art) capture the **genre atmosphere** named in the instruction (e.g., mythic epic, sci-fi)?\n- Is the **main character prominently featured** with a recognizable likeness from the reference and a compelling hero pose/lighting?\n- Are professional **poster conventions** followed (title treatment, tagline, billing-block area, focal composition)?\n- Do color palette, lighting and typography reinforce the intended mood and attract the target audience?\n- Is the title text rendered clearly and integrated into the composition.",
+    "IP与娱乐-皮肤装备": "- Does the redesigned **skin / equipment set authentically embody the specified theme** (e.g., gothic-dark stage skin, close-combat gear) while keeping the character's base silhouette recognizable?\n- Is the equipment **functionally appropriate to the character's implied class/role** (a brawler gets light close-combat gear, not heavy siege armor)?\n- Are theme-specific materials, textures and decorative details rich and internally consistent?\n- Does it read as premium, collectible-quality design rather than a flat recolor?\n- Is the character's core identity/likeness preserved through the reskin.",
+    "地理与历史-历史实体信息化呈现": "- Does the generated artifact (e.g., a historical figure's resume/profile for a modern role) **accurately reflect the figure's real achievements, era and expertise**, reframed for the requested modern position?\n- Is the **mapping between historical accomplishments and the modern job requirements** clever and logically justified (e.g., a strategist → operations director)?\n- Are biographical facts (dates, titles, deeds) historically correct?\n- Is the information presented with professional document structure and clear hierarchy?\n- Is any portrait consistent with the reference figure.",
+    "地理与历史-地图渲染": "- Does the rendered ground-level / landscape scene **faithfully reflect the geographic features encoded in the source map** (street layout, terrain/contour lines, coastline, elevation)?\n- For contour/topographic inputs: is the **terrain elevation and landform** correctly interpreted into 3D relief (peaks, valleys, slopes)?\n- For street-map inputs: does the real-scene view match the road network, block density and landmarks of that location?\n- Is the atmosphere/time-of-day as requested, with physically plausible lighting?\n- Are local architectural and environmental characteristics of the region authentically depicted.",
+    "地理与历史-地图路线": "- Is the drawn route **geographically valid on the given map** — following actual roads/paths, connecting the specified start, waypoints and destination in correct order?\n- Are all **named waypoints included and correctly located** (e.g., specific stadiums, buildings, parks, stations)?\n- Is the route mode-appropriate (driving vs. cycling vs. walking uses sensible roads) and drawn as a clear highlighted line with direction?\n- Are start/end and key stops clearly marked/labeled?\n- Does the annotated map remain readable (route not obscuring essential map detail).",
+    "影视与叙事-动画关键帧": "- Do the keyframes depict a **coherent motion sequence** (e.g., run→sprint, a dance passage) with correct pose progression and a clear sense of movement/speed?\n- Is the **exact number of frames** produced as requested, in correct temporal order?\n- Are **character design, costume and body proportions consistent across all frames** (continuity)?\n- Do the poses obey anatomy and physics (weight, balance, momentum) rather than arbitrary stances?\n- For narrative frames (e.g., 昭君出塞): do the frames convey the intended story beats.",
+    "影视与叙事-影视概念图": "- Does the concept art establish a **coherent film mood/genre** as described (e.g., nostalgic coming-of-age, uncanny mystery)?\n- If **two reference images are to be combined**, are both meaningfully integrated into one unified scene rather than pasted side by side?\n- Does the composition convey the narrative premise (the described place/situation and its emotional undertone)?\n- Are lighting, color grading and set design consistent with the target genre?\n- Is it cinematic and atmospheric (production-concept quality), not a plain snapshot.",
+    "影视与叙事-绘画过程": "- Does the sequence show a **believable art-creation progression** for the given medium (e.g., charcoal underdrawing → blocking → shading → finished portrait; sketch → lineart → color → finish)?\n- Is the **specified number of steps** produced, in correct order, each step a distinct, meaningful stage?\n- Does the **final stage match the referenced artwork/subject**, and do intermediate stages plausibly lead to it (identity/composition preserved throughout)?\n- Are the techniques appropriate to the medium (pencil/charcoal/oil/digital)?\n- Is the progression from rough to refined visually clear.",
+    "数据与信息-电商产品图": "- Does the e-commerce hero/main image **accurately reflect the product's key specs and selling points** from the provided description (capacity, features, target users, brand positioning)?\n- Is the **product itself the clear focal point**, faithfully preserved from the reference (shape, color, form factor) with a clean, attractive presentation?\n- Are **feature callouts / spec highlights** laid out with clear visual hierarchy and correct, legible text?\n- Does the visual style match the brand positioning and appeal to the stated target audience?\n- Does it follow professional e-commerce main-image conventions (hero shot, benefit highlights, brand cues).",
+    "时尚与美妆-专业服装": "- Is the outfit **functionally correct for the specified activity/sport** (e.g., skiing → insulated waterproof layers, goggles; badminton → light breathable sportswear, proper shoes)?\n- Are **activity-specific gear and safety/performance elements** included and appropriate?\n- Does it consider fit, freedom of movement and climate suitability for that activity?\n- Is the person's identity/body preserved while only the outfit changes?\n- Is the result realistic and wearable rather than a costume caricature.",
+    "时尚与美妆-发型发色": "- Does the hairstyle/hair-color design **address the stated concern or goal** (e.g., add apparent volume for flat hair; a retro updo look) with a realistic, achievable result?\n- Is it **suited to the person's hair length, face shape and features**, and to any stated constraints (e.g., not overly trendy, medium-short hair)?\n- For multiple recommendations: are the options **diverse and clearly distinct** and the requested count produced?\n- Are the person's facial features and skin tone preserved (only hair changes)?\n- Is the hair rendered naturally (texture, hairline, color transition believable).",
+    "时尚与美妆-场合造型": "- Is the complete look **appropriate to the specified occasion** (e.g., traditional Chinese wedding, chemistry-lab work) covering clothing plus relevant hair/makeup/accessories where implied?\n- Are **occasion- and culture-specific conventions and safety requirements** respected (e.g., lab: closed shoes, tied hair, coat; wedding: proper ceremonial attire)?\n- Is the styling coherent (colors, formality, silhouette all consistent with the setting)?\n- Is the person's identity preserved while the styling changes?\n- Is the result realistic and contextually convincing.",
+    "时尚与美妆-妆容诊断": "- Does the diagnostic image **correctly identify the specific problem areas** named (e.g., base, brows, eye makeup, blush, lip balance) with clear visual markers/annotations pointing to the right regions?\n- Are the **critiques accurate and the suggested corrections professionally sound** and natural-looking?\n- Does it present a clear before/annotation (and improved direction where asked) that is easy to follow?\n- Are annotations legible and correctly placed?\n- Is the person's facial identity preserved.",
+    "时尚与美妆-护肤美妆产品": "- Does the guide **correctly use the specific products shown** (e.g., an eyeshadow palette's base/deepen/highlight shades; a set of cosmetics) and explain **what each product is for and where to apply it**?\n- Is the **application order and placement professionally correct** for the stated goal (daily eye look, beginner commute makeup)?\n- Are steps clearly laid out with legible labels mapping products to face regions?\n- Is the resulting look natural and suitable for the stated scenario?\n- Are the depicted products consistent with the reference.",
+    "时尚与美妆-穿搭分析": "- Does the analysis **correctly diagnose the stated problem** (e.g., why an outfit looks short/unbalanced) and produce an **improved version that visibly fixes it** while honoring constraints (e.g., keep a specific garment)?\n- Are the critiques specific and actionable, shown with clear annotations or a before/after comparison?\n- Is the improved styling coherent (proportion, color, silhouette) and realistic?\n- Is the person's identity preserved across the analysis/redo?\n- Are any retained items faithfully kept.",
+    "生活与消费-园艺宠物": "- Is the gardening/pet plan **horticulturally/behaviorally sound and practical** for the given space (light exposure, watering needs, planting layout, care cycle; or pet zones and needs)?\n- Are plants/animals matched to the space's real conditions, with correct grouping (sun-loving vs. frequent-watering, etc.)?\n- Are layout, placement and a care/maintenance schedule clearly shown and implementable?\n- Is the original space structure preserved while only the arrangement changes?\n- Are annotations/labels clear and accurate.",
+    "生活与消费-室内渲染": "- Does the interior rendering **satisfy the stated living/use requirements** (e.g., occasional small concert yet normal daily living; a small B&B with reception/rooms/lounge/circulation)?\n- Is the layout **spatially and architecturally plausible** given the original room/CAD structure (walls, dimensions, floor relations preserved)?\n- Are zoning and circulation logical, with furniture/fixtures placed sensibly for the stated function?\n- Is the design style coherent and the render realistic (proportion, lighting, materials)?\n- Does it keep the original structural constraints while re-planning the interior.",
+    "生活与消费-家庭安全与维护": "- Does the image **correctly identify real safety hazards / maintenance issues** for the scene (e.g., kitchen/stove risks) and give **professionally correct cleaning or fixing steps and materials**?\n- Are **hazard points accurately marked** and the recommended safe practices genuinely safer?\n- For cleaning tasks: are the tools/products and step order practical and correct, with sensible cautions?\n- Is the room/appliance faithfully preserved while annotations/plans are overlaid?\n- Are annotations and steps legible and well organized.",
+    "生活与消费-居家健身": "- Is the home-workout plan **physiologically sound and appropriate to the target user** (e.g., desk workers, seniors) and to the space and constraints (e.g., low-noise for neighbors)?\n- Does it include a **sensible structure** (warm-up → main/strength → stretch/cool-down) with correct, safe exercises and reasonable duration?\n- Are exercises feasible within the shown space and equipment-free (or minimal) as implied?\n- Is the plan clearly presented (steps/zones/illustrations) and easy to follow?\n- Is the original space preserved as the workout setting.",
+    "生活与消费-活动布置": "- Does the event-decoration plan **fit the specified occasion and theme** (e.g., cream-toned birthday party, wedding anniversary) with cohesive, on-theme decor?\n- Are **practical event elements considered** (photo backdrop, cake/food table, dining setup, guest circulation, easy cleanup) as relevant?\n- Is the styling coherent (color scheme, mood, density) and realistically achievable in the given room?\n- Is the original room structure preserved while decor is added?\n- Does it look like a professionally styled event rather than random props.",
+    "生活与消费-饮食健康": "- Is the nutrition/meal artifact **factually sound** (e.g., a food's farm-to-table journey is realistic; a multi-day meal plan is balanced and varied)?\n- For meal plans: does it **reuse ingredients sensibly, avoid repetition, and flag extra items to buy**, respecting the stated household/constraints?\n- For process/origin graphics: is the sequence correct and the required panel count produced?\n- Is the information presented clearly (grid/steps/labels) and easy to act on?\n- Are ingredients/dishes recognizable and consistent with any reference.",
+    "科学与教育-常识推理": "- Does the result **correctly predict the physical/temporal outcome** implied by the instruction (e.g., how an object looks after sitting at room temperature for an hour; a foil balloon deflating and wrinkling after a week)?\n- Is the change **physically plausible in direction and degree** (melting, wilting, oxidation, deflation, condensation, etc.), not exaggerated or wrong?\n- Are elements not affected by the process kept consistent with the reference image?\n- Is the depicted state realistic and clearly showing the reasoned change?\n- Does it reflect genuine everyday-physics/commonsense understanding.",
+    "设计与产品-包装VI设计": "- Does the packaging/label design **accurately reflect the product's brand, spec and selling points** from the brief (name, variant, volume, positioning, key claims like 0-sugar)?\n- Is the **brand tone and target mood** correctly expressed (e.g., calm/premium for a relaxation candle; bright/refreshing for a lemon soda) via color, typography and layout?\n- Does it follow professional packaging/label conventions (product name prominence, required info, balanced composition) with legible, correct text?\n- Is the original product/bottle form preserved where required?\n- Is it cohesive across pieces (label + box) and production-ready in appearance.",
+    "数据与信息-数据分析解读": "- Does the output go beyond re-plotting the raw table and actually **derive analytical insight** required by the instruction (e.g., top-N ranking with drivers, radar comparison of multi-dimensional ability, sales-vs-turnover distribution, attrition-risk prediction & attribution)?\n- Are the **numbers/relationships faithful to the source data** in the input image (rankings, magnitudes, comparisons are correct, not fabricated)?\n- Is the **chart type appropriate to the analytical question** (ranking→bar/leaderboard, multi-dimension comparison→radar, distribution→scatter/bar, driver/attribution→annotated breakdown)?\n- Are conclusions/annotations (e.g., \"who is strongest\", \"which items are slow-moving\", \"high-risk employees\") **supported by the shown data** and clearly highlighted?\n- Are axes, legends, labels and key takeaways legible and correctly placed, telling a clear data story.",
+    "时尚与美妆-化妆教学图解生成": "- Does the tutorial correctly analyze the person's **actual facial features** (eye shape, face contour, lip shape) and give techniques **tailored to them** (e.g., de-puff/deepen for a specific eye type, contour&highlight placement for a face shape)?\n- Are the **application steps in correct order and each step visually distinct**, showing progressive buildup (base → eyes → contour/highlight → lips) as relevant?\n- For before/after or effect-comparison requests: is the transformation realistic and does it preserve the person's identity?\n- For multi-style variants (e.g., three lip styles): are the styles distinct and correctly characterized?\n- Are step labels/annotations legible and the makeup techniques professionally accurate.",
+    "时尚与美妆-服饰文化": "- Does the infographic **accurately explain the cultural symbolism and meaning** of the depicted traditional garment (motifs, colors, components and what they signify), not just describe its appearance?\n- Are the **historical period, ethnic/regional origin and social-cultural context** factually correct?\n- Are individual garment parts **correctly identified and annotated** with their cultural/symbolic function?\n- Is the information organized as a clear, readable explanatory graphic (labels, callouts, hierarchy) that faithfully references the garment in the image?\n- Is the cultural interpretation authentic and free of anachronism or fabrication.",
+    "生活与消费-画像生成": "- Does the generated portrait depict a **plausible matching partner that logically satisfies the stated need** (e.g., a sociable companion, a home-oriented/caretaking partner, a fellow adventurer, someone good with children)?\n- Are the person's **implied traits reflected as coherent visual cues** (age range, temperament, lifestyle, style) consistent with the described compatibility goal?\n- Is the reasoning defensible — the portrait's characteristics can be read as fitting the requested trait, rather than arbitrary?\n- Is it a realistic, high-quality single-person portrait (not a collage), appropriate to the context implied by the reference person?\n- Does it avoid contradicting the given constraints.",
+    "生活与消费-餐饮美食场景化": "- Does the generated dining scene **match the occasion/theme derived from the menu** (e.g., New-Year family reunion, snowy-night hotpot gathering, high-end business banquet, romantic candlelit dinner)?\n- Are the **dishes on the menu plausibly represented on the table**, consistent with the cuisine and setting?\n- Does the ambiance (lighting, decor, table setting, guest arrangement) authentically convey the intended mood and formality level?\n- Is the scene coherent and realistic (spatial layout, scale, props all fitting the occasion)?\n- Are menu-implied specifics (number of diners, cuisine type, festivity) respected.",
+    "科学与教育-排版重构": "- Does the reconstruction **restructure the source document to serve the stated goal** (e.g., split a thick chapter into N study-card modules; turn an official policy into a one-page plain-language guide; flag abnormal medical values with advice; condense scattered financials into a radar chart)?\n- Is the **key information preserved and correctly reorganized** — no critical content lost or distorted, correct grouping into the requested structure/module count?\n- For data-to-chart requests: is the chosen chart type appropriate and the values faithful to the source?\n- For annotation requests (e.g., mark abnormal items, add advice): are the flagged items correct and the added guidance sound?\n- Is the new layout clear, legible and genuinely easier to read than the original.",
+    "科学与教育-逻辑推演": "- Does the answer **correctly solve the visual reasoning puzzle** (e.g., infer the missing pattern in the '?' cell of a matrix/sequence)?\n- Is the **underlying rule correctly identified** (rotation, progression, arithmetic on shapes, symmetry, set operations, etc.) and consistently applied?\n- Does the produced answer **fit the puzzle's visual style** (same shapes, line weight, arrangement) and contain exactly the correct content — no extra or missing elements?\n- Is the result unambiguous and logically forced by the given premises?\n- Is the drawn answer clean and clearly placed in the target cell.",
+    "设计与产品-产品实图化": "- Does the output **faithfully turn the sketch/line-art/design draft into a realistic product image** while preserving the design's form, proportions and key features?\n- Are the requested **materials and surface qualities accurately rendered** (e.g., matte metal, plastic gloss, diamond & gold sparkle, injection-mold parting lines/mold marks)?\n- Is the rendering physically plausible (lighting, reflections, shadows, edges consistent with the stated material)?\n- If a scene/context is requested (e.g., toy in a child's room), is the product correctly integrated at the right scale with realistic interaction?\n- Does it read as a professional product photo/render rather than a flat repaint of the sketch."
+  }
+}