Core Concept
Through the technique of juxtaposition, reveal the universal dilemma of the "huge gap between promise and delivery" in any field. The left side presents an idealized perfect render, while the right side reveals the real everyday appearance.
Execution Flow
- Receive image → Analyze content, identify the field it belongs to as
$domain, calculate the image aspect ratio as$ratio - State determination → Determine the current state (dilapidated/ordinary/ideal)
- Intent recognition → Determine the output mode according to the user's instructions
- Parameter mapping → Map the general five dimensions to field-specific expressions
- Build prompt → Build a prompt based on field-specific expressions
- Generate image → Generate the target image
works well with skills: image-to-prompt, prompt-to-image
Working Modes
1. Image State Recognition
After the user uploads an image, analyze its current state:
| State | Features | Output target |
|---|---|---|
| Dilapidated | Quality issues, signs of use, poor maintenance | Generate an idealized render (i.e., step 2.1) |
| Ideal | The user uploaded marketing images, advertising images, or other retouched photos | Generate a slightly dilapidated render image (i.e., step 2.2) |
| Ordinary/Normal | No obvious damage, everyday use state | Generate a comparison image of an idealized render and a slightly dilapidated version (i.e., step 2.3) |
If the state is ambiguous and the intent cannot be determined, proactively ask the user for the desired direction
2. Three Output Modes
2.1 Idealized Render (Ideal)
- Goal: Output a promotional-level perfect presentation for the corresponding field
- Features: High saturation, perfect lighting and shadow, flawless, carefully composed
2.2 Real Appearance (Reality)
- Goal: Output the real everyday state of the corresponding field (not dilapidated)
- Features: Natural light ("death lighting"), realistic texture, everyday atmosphere, unretouched
2.3 Comparison Image (Comparison)
- Goal: Output a comparison image that juxtaposes an idealized render and the real appearance
- Arrangement: Automatically choose left-right or top-bottom arrangement based on the original image's aspect ratio
Field Recognition and Adaptation
Field Detection
Based on image content keyword matching:
建筑领域: 建筑外观、城市景观、室内空间、建筑效果图、楼盘、住宅、商业空间
人像领域: 人像写真、Cosplay、自拍、证件照、活动拍摄、肖像
产品领域: 电商产品、商品展示、包装设计、电子产品、服饰
食物领域: 美食摄影、菜品展示、烘焙、饮品、餐厅菜单
旅游领域: 风景照、景点打卡、酒店房间、度假胜地
游戏领域: 游戏截图、游戏宣传、UI界面、角色设计
健身领域: 健身照、运动场景、瑜伽、健身房
家居领域: 室内装修、家具展示、样板间、智能家居
科技领域: 产品发布会、概念设计、VR/AR、智能汽车
Core Comparison Dimensions (General Framework)
All fields share the following five core comparison dimensions:
1. Lighting
| Ideal side | Reality side |
|---|---|
| Carefully calculated perfect lighting | Natural/actual on-site lighting |
| Golden hour or soft fill light | Hard light, top light, or flat diffused light |
| Rich tonal gradation, no crushed blacks/overexposure | Exposure compromises, heavy shadows |
| Clear directionality, strong sense of volume | Low contrast, lacking depth |
2. Material
| Ideal side | Realistic side |
|---|---|
| Flawless surface | Signs of real use |
| Saturated colors, enhanced texture | Fading, stains, wear |
| No dust, no water marks, no flaws | Natural aging, environmental traces |
| CG-like precise reflection/refraction | Murky, imperfect reflections |
3. Color
| Ideal side | Realistic side |
|---|---|
| High saturation, vivid and eye-catching | Low saturation, somewhat flat |
| Accurate color temperature, unified and harmonious | Color temperature shift, uncorrected white balance |
| Color enhancement from post-processing | Native camera color reproduction |
| Advertising-grade visual appeal | Everyday feel, plain feel |
4. Atmosphere
| Ideal side | Realistic side |
|---|---|
| Full of energy, vibrant | Cold, ordinary, or slightly awkward |
| Carefully arranged scene elements | Cluttered on-site environment |
| Dreamy, idealized background | Real environment that exposes the scene |
| Emotionally rich, captivating | Documentary feel, cold and objective |
5. Composition/Details
| Ideal side | Realistic side |
|---|---|
| Perfect perspective and proportions | Natural lens distortion |
| Flaw removal, continuity error fixes | Retain all on-site details |
| Carefully arranged element layout | Random, irregular real distribution |
| Post-added special effects/light effects | Original state without post-processing |
Comparison Image Technical Specifications
Arrangement Rules
If generating a comparison image, determine the new arrangement rules based on the original image's aspect ratio $ratio :
- Original image horizontal(width > height): comparison image arranged top and bottom, horizontal dividing line
- Original image vertical(height > width): comparison image arranged left and right, vertical dividing line
Dividing Line Specifications
- Position: exact center of the frame
- Width: 2-5 pixels
- Color: pure white or very light gray
- Edges: sharp and clear, no feathering
Content Rules
- Left or top of the comparison image (according to the arrangement rules): idealized rendering
- Right or bottom of the comparison image (according to the arrangement rules): ordinary realism (default) or ruined state (when the user explicitly requests a strong contrast effect)
- By default, do not add title text, keep the image clean, and let the visual contrast itself speak