Anti-Render Intelligent Image Transformation

Task Objectives

• This Skill is used to: intelligently identify the field of the image uploaded by the user, and generate images in corresponding styles based on the core concept of "Ideal Promise vs. Harsh Reality"
• Capabilities include: domain recognition (architecture/person/others), image status recognition, intelligent intention judgment, three-mode image generation (ideal rendering/true appearance/comparison image)
• Trigger conditions: The user uploads an image from any field and requests "anti-render processing", "make a comparison", "idealization processing" or "realistic processing"

Preparations

• No special dependencies required, utilizing the existing image recognition and generation capabilities of the agent
• The user needs to provide the image to be processed
• According to the recognized domain, dynamically read the corresponding style feature library, and extract inspiration and style key points from JSON (instead of following or copying the text verbatim):
  • Architecture domain: anti-render.architecture.json
  • Person domain: anti-render.person.json
  • Other domains: Use general core principles

Work Steps (Intention Recognition)

• Architecture domain: cracked walls, peeling paint, rusted metal
• Person domain: skin problems, poor lighting, bad makeup
• Other domains: obvious quality issues, usage traces, poor maintenance
• Output Target: Return an idealized rendering that reaches the promotional level of the field

• Architecture domain: normal building appearance, no obvious damage, daily usage traces
• Person domain: normal camera output, no serious skin problems
• Other domains: no serious quality issues, normal usage status
• Output Target: Return a corresponding realistic "true appearance" image

• The user explicitly uses the keyword "comparison"
• Output Target: Return a comparison image, split by a prominent but non-distracting dividing line in the middle
• Arrangement method: Determine whether to arrange vertically or horizontally according to the aspect ratio of the original image, so as to return a new image with a more balanced aspect ratio
  • Horizontal image (width > height): Arrange left and right, with a vertical dividing line
  • Vertical image (height > width): Arrange top and bottom, with a horizontal dividing line
• Default Comparison: Extreme Ideal vs. Ordinary Reality (unless the user explicitly requests "dilapidation comparison")

• The image status is between damaged and ordinary, and the user's intention cannot be clearly judged
• Action: Need to actively ask the user for the desired transformation direction

Core Concept

• Advertising Image vs. Actual Product: Polished promotional images of e-commerce products vs. real photos from buyers
• Concept Design vs. Mass-Produced Version: Concept images of cars/phones at press conferences vs. the compromised final version on the market
• Packaging Design vs. Shelf Product: Exquisite packaging in design drafts vs. actual effects on supermarket shelves
• Menu Photo vs. Served Dish: Exquisite plating on restaurant menus vs. the real dish served by waiters
• Food Advertisement vs. Unboxed Product: Tempting close-up of a hamburger in ads vs. the collapsed hamburger after unwrapping
• Baking Tutorial vs. Failed Attempt: Perfect finished product from a food blogger vs. a beginner's first try
• Official Promotional Photo vs. Tourist's Shot: Empty beautiful scenery from tourism bureaus vs. the real crowded scene
• Hotel Official Website Image vs. In-Room Shot: Showroom of a five-star hotel vs. the actual room assigned
• Scenic Area Rendering vs. Completed Scene: Planned theme park vs. the cheap feeling after opening
• Urban Planning Map vs. Actual Construction: Future city vision displayed by the government vs. the actual appearance after ten years
• Public Space Design vs. Current Usage: Vibrant square in the designer's vision vs. the empty cement space that no one cares about
• Traffic Planning Map vs. Congested Reality: Idealized road flow simulation vs. the parking lot-like morning rush hour
• Game Promotional Video vs. Actual Gameplay: Polished demo at E3 vs. actual running effect on players' computers
• Movie Poster vs. Screenshot: Polished character poster vs. actual shot in the movie
• Game UI Concept vs. Final Interface: Cool interface in design drafts vs. the simplified version after launch
• Fitness Advertisement vs. Real Training: Perfect figure and environment promoted by gyms vs. the real sweaty training scene
• Weight Loss Before-After vs. Real Process: Perfect transformation on social media vs. the repeated struggles in the middle
• Yoga Tutorial Image vs. Beginner's Shot: Perfect pose from the instructor vs. the stiff imitation of an ordinary person
• Co-working Space Promotion vs. Actual Usage: Ideal WeWork-style office space vs. the crowded and noisy real environment
• Home Showroom vs. Post-Move-In: Perfect storage in IKEA showrooms vs. the mess after three months of living
• Smart Home Demo vs. Daily Usage: Smooth operation at tech exhibitions vs. the frequent disconnection at home
• Music Festival Official Announcement vs. Live Scene: Cool stage design from the organizer vs. the muddy and crowded audience view
• Wedding Rendering vs. Ceremony Scene: Dreamy arrangement from wedding companies vs. the actual effect with limited budget
• Exhibition Rendering vs. Post-Exhibition Scene: Exquisite booth setup at the exhibition opening vs. the dilapidated scene on the last day
• Course Promotional Image vs. Actual Classroom: Exquisite courseware from online education vs. the lagging live stream
• Campus Promotional Video vs. Student's Daily Life: Ideal campus life in admission brochures vs. the real scene of fighting for seats in the library
• Training Institution's Promise vs. Employment Reality: High-paying employment promoted vs. the actual job-hunting dilemma
• Environmental Protection Promotion vs. Pollution Status: Green mountains and clear waters in government reports vs. actual industrial pollution
• Seasonal Promotional Photo vs. Climate Reality: Picturesque four seasons in travel brochures vs. abnormal weather due to global warming
• Wildlife Documentary vs. Habitat Status: BBC-level magnificent nature vs. the desolation after human activity destruction
• Instagram Life vs. Real Daily: Carefully curated lifestyle photos vs. the messy room outside the camera
• Internet Celebrity Check-in Spot vs. Beyond the Shooting Angle: Perfectly composed internet celebrity wall vs. the truth of a garbage dump just around the corner
• Dating App Photo vs. In-Person Meeting: Polished profile photos vs. the real appearance at the first date
• Press Conference Rendering vs. Mass-Produced Device: Perfect industrial design at Apple's press conference vs. the actual antenna band/notch
• VR Experience Promotion vs. Actual Wearing: Immersive future in ads vs. the dizziness and bulkiness after putting on the headset
• Smart Car Demo vs. Road Test Reality: Ideal autonomous driving demo vs. the frequent takeover during testing

Domain Recognition and Resource Mapping

1. Architecture Domain

2. Person Domain

3. Other Domains

Core Requirements

• The complete structure and all elements of the original image must be strictly retained in all modes
• Ideal Rendering/True Appearance mode: The output image maintains the original size (W×H)
• Comparison Image mode: The left/right or top/bottom parts each maintain the original size and composition of the image (each part is W×H), only changing materials, lighting, colors and atmosphere; the final canvas is 2W×H or W×2H
• The dividing line should not damage or block the core structure of the original image
• Do not change the composition, perspective, key forms, environmental layout, etc. of the original image
• Prohibit adding any new elements such as titles, borders: No decorative elements that are not in the original image should be added to the output image, including but not limited to titles, text labels, borders, decorative lines, watermarks, etc.

• Intelligently determine the arrangement method according to the aspect ratio of the original image
• Obtain a new image with a more balanced aspect ratio through splicing
• Dividing line position: centered, 2-5 pixels wide, pure white or very light gray
• Edge processing: sharp and clear, no feathering, emphasizing the sense of fracture between the two worlds
• The dividing line itself is not a new element, but part of the composition

• Default Comparison: Extreme Ideal vs. Ordinary Reality
  • Left/Top: Extreme idealized rendering (perfect presentation of the field)
  • Right/Bottom: Ordinary realistic appearance (real daily state, but not dilapidated)
• Special Comparison: If the user explicitly requests "dilapidation comparison" or "extreme comparison", use Ideal vs. Dilapidated comparison
  • Right/Bottom: Harsh reality (serious quality problems, decay, dilapidated state)

• Architecture Domain: Must read anti-render.architecture.json, extract key points of color, lighting, material, atmosphere, lens feel, etc. to guide generation; do not copy the original sentences of JSON or treat the text as elements to be generated in the image
• Person Domain: Must read anti-render.person.json, extract key points of skin texture processing, lighting method, color grading tendency, background texture, etc. to guide generation; do not copy the original sentences of JSON or treat the text as elements to be generated in the image
• Other Domains: Use general core principles, create based on 5 core comparison scenarios

• If the user's image is too neutral and the user's intention cannot be identified, need to actively ask the user
• The inquiry content clearly lists three transformation directions for the user to choose