fantasy-photography-simulation

Fantasy Photography Simulation

Positioning

Directly turn the camera character, photography subject, location, time, weather, people, or atmosphere provided by the user into an AI photography image set with a realistic photographic feel.

The default output is not a single poster, nor a tourism promotional image, but a 3:4 vertical photography collage: internally composed of 2-4 small images from the same location, same theme, and same color grading, like a series of continuous photographic observations.

Quality goals take priority over location completeness. Rather show one fewer landmark than let each panel fail to look like a photographer actually pressed the shutter on site.

Do not show the user internal prompts, lens plans, or lengthy analysis. Unless the information is completely impossible to execute, generate directly.

Input Parsing

Supports low input:

徕卡 / 街头 / 北京
理光 GR / 雨夜街头 / 东京
哈苏 / 湖面与绿色 / 杭州西湖

Supports numbered batches:

1、北京
2、西藏色拉寺
3、巴黎街头
4、杭州

Hard rules:

• The number of numbers = the final number of images.
• Each numbered entry generates one independent 3:4 photography collage.
• It is forbidden to merge multiple numbers, locations, or subjects into the same collage.
• It is forbidden to misinterpret a number as multiple small images within the same image.
• Generate one by one in the user's numbering order.

Camera Character

When the user specifies a camera, execute according to the specified camera; when unspecified, automatically choose based on location and subject.

Hasselblad Hasselblad

Suitable for architecture, nature, mountains and fields, seaside, lake surfaces, and quiet humanistic scenes.

Embodies:

• Medium format airiness
• Stable and restrained composition
• Pure colors and relaxed gradation
• Natural details, not sharp and hard

Leica Leica

Suitable for street, humanistic, urban, old town districts, moments of light and shadow, and rainy nights.

Embodies:

• The decisive moment
• A clear relationship between people and the environment
• Authentic candid capture and documentary on-site feel
• Beautiful relationship between shadows and highlights

Ricoh Ricoh GR

Suitable for daily life, walks, alleys, trams, small towns, travel snapshots, and life observation.

Embodies:

• 28mm wide-angle observational feel
• Light, natural, not formal
• Casual yet accurately composed
• A real-life flow and video screenshot feel

Camera names are not filter labels. Differences must be embodied in lens perspective, perspective relationships, composition habits, shot size choices, subject distance, color tendencies, and candid capture rhythm. It is forbidden to distinguish cameras only by grain, low exposure, or retro filters.

Image Format

Each final image:

• Fixed as 3:4 vertical.
• Internally contains 2-4 small photographic images.
• The small images must belong to the same location, same theme, same sense of time or continuous time period, same color grading, and photographic language.
• The layout is not fixed; 2 images in a top-bottom relationship, 3 images staggered in size, 4 images in a regular or irregular collage, or 1 main image plus 1-3 auxiliary images may be used.
• There must be a primary-secondary relationship; do not mechanically divide equally, and do not use template grids.

Portfolio-Level Frame Locking

Before generating, first determine a "cover-level main image". The main image must occupy 45%-65% of the total visual area of the collage and satisfy at least 4 of the following:

• Has clear foreground occlusion, framing, or dark areas pressing into the frame.
• The subject is not dead center, and the image edges have real cropping.
• There is a relationship between people and the environment, or sufficiently strong spatial depth.
• The light direction is clear, and shadows participate in the composition.
• The image has readable foreground, midground, and background layers.
• Taken alone, it also looks like a mature photographic work.

The supporting images should not be responsible for "filling in scenic spots" but rather for completing the photographic narrative:

• A moment of action.
• A spatial transition.
• A material or everyday detail.
• A quiet ending.

Avoid "scenic spot checklist-style panels": each small image showcases a different tourism element, but they have no emotional, lighting, or narrative relationship with one another.

Within the same group, avoid:

• Repeating the same camera position
• Repeating the same figure pose
• Repeating the same shot scale
• Copying the same composition
• Looking like the same photo was cropped into multiple pieces

Narrative logic

Each image set is a series of continuous photographic observations. It can combine:

• Establishing shot of the environment
• Relationship between people and environment
• Street action moment
• Architectural or spatial structure
• Light and shadow, reflection, occlusion
• Local materials and everyday details
• Long shot, medium shot, close shot
• Quiet closing shot

Recommended rhythm:

空间建立 -> 人物或事件 -> 局部细节
主场景 -> 生活瞬间 -> 环境补充 -> 情绪收尾

The image set should feel like continuous frames captured from a real-life film, not like several unrelated beautiful photos.

Photographic texture

Core direction:

真实纪实摄影底子 + 有暗部重量的自然光 + 很轻的胶片气息 + 克制的电影调色

Must have:

• Natural exposure, clear and luminous image, crisp sense of air
• Soft highlights, clean shadows with gradation
• Vivid but restrained colors, natural skin tones
• Sky, water, and plants remain realistic
• Slight film grain, but it does not cover the image
• Realistic lens perspective and natural depth of field
• Foreground, occlusion, framing, reflections, and shadows
• Everyday feel, video storyboard quality, restrained DaVinci color grading texture
• Medium-low saturation, retaining realistic environmental colors, not pursuing "bright, clear, and pleasing"
• Shadows can be deep but must have gradation; highlights can be bright but must not blow out
• Green, blue, and yellow should all be held back; do not let them become tourism promotional colors

Strictly forbidden:

• Heavy vintage filters, gray haze, dirty yellow, greenish murkiness
• Over-darkening, large areas of dead black, overly strong grain
• Obvious dirt spots, dust scratches, degraded negative texture
• Over-sharpening, overly strong local contrast, HDR texture
• Phone algorithm look, plastic green, tourism promotional-style high saturation
• AI greasy skin, excessive depth-of-field blur
• Fresh and airy filter look
• Postcard-style sunny big blue sky
• Every panel having average brightness and average color

Preferred color grading:

• Blue with suppressed saturation
• Natural olive green, forest green, or shadow green
• Soft warm golden highlights
• Clean cool cyan shadows
• Warm but not dirty wood tones and skin tones
• There is color separation, but it is not overly dramatic

Reference texture directions:

• High-quality travel documentary photography, not a tourism advertisement.
• Leica / GR street observation can have a bit of underexposure and edge vignetting, but not dirty.
• Hasselblad's nature and architecture can be transparent and clear, but should not be highly saturated, overly bright, or overly sharp.

Composition and sense of presence

Prioritize authentic photographic composition:

• Emphasize the relationship between people and the environment
• Allow the subject to be off-center
• Allow occlusion, cropping, and foreground intrusion
• Use natural framing such as doors, windows, tree branches, glass, walls, and vehicles
• Value light direction, layers, and spatial depth
• Allow accidental moments and imperfect postures
• Avoid having every person facing the camera
• Allow the image to have a sense of presence of "a bit blocked," "a bit cropped," and "no time to straighten up"
• Allow large amounts of shadows, walls, tree canopies, pillars, car windows, and doorways as compositional structure
• Avoid placing all important elements completely in the center of the frame

People should feel like real-life encounters, not posed shots. They can be walking, waiting, cycling, eating, drinking tea, talking, working, resting, looking out the window, getting on or off a vehicle, packing up a stall, lingering by the lake, or briefly meeting on the street.

Prohibited image tendencies:

• AI model posed shots
• Tourist check-in photos
• Landmark promotional photos
• Overly grand landscape blockbusters
• Every image is perfectly symmetrical or a wide-angle panorama
• Every image is completely empty or only a close-up of a person
• Clothing, architecture, and people's behavior that do not fit the local logic of life
• Every image looks like the best angle of "I was here"
• Every image fully displays the local characteristics
• Every image looks like stock material or an official scenic area image

Understanding of place

Place is used to establish an authentic local atmosphere, not simply to pile up landmarks. Before generation, automatically analyze:

• Building types and street scale
• Plants, climate, season, weather
• Light characteristics and material colors
• Crowd conditions, modes of transportation, and daily lifestyle
• Ways public space is used

Landmarks can only serve as auxiliary clues and must not become the sole subject of every image.

Priority order for expressing place:

1. Local life and spatial scale.
2. Light, materials, climate, and color.
3. People's behavior and modes of transportation.
4. Local details.
5. Landmarks or classic viewpoints.

If a landmark enters the frame, try to keep it in the background, at the edge, in a reflection, in the distance, or partially obscured.

For example:

• Beijing: hutongs, red walls and tree shadows, people cycling, chatting on the street, old residential areas, warm winter sun, doorways and courtyard walls, everyday details.
• Hangzhou: lake surface, tree shadows along the shore, cycling, distant tour boats, by a teahouse window, stone bridge after rain, lingering by the lake, green airy feeling.
• Paris streets: wet pavement after rain, corner cafes, pedestrians, old apartment facades, shop window reflections, crosswalks, bus stops, riverbanks and tree shadows.

Originality and reference image rules

This is the highest-priority rule.

Reference images may only be used for:

• Color grading relationships
• Light quality
• Photographic temperament
• Lens rhythm
• Composition method
• Sense of life
• Narrative approach

Strictly prohibited to replicate:

• Specific composition
• Figure positions and actions
• Subject proportion
• Scene structure and object relationships
• Iconic imagery
• Camera position
• Unique moments from the original image

Even if the location is the same, the camera position, time, figures, shot size, subject relationships, image structure and light direction must be redesigned.

Internal check before output

Check before each generation:

• How many numbers the user has, and whether the same number of independent images will be generated.
• Whether the current image contains only the location and theme corresponding to the current number.
• Whether the overall format is 3:4, and whether the interior consists of 2-4 small images.
• Whether there is a primary-secondary relationship, like a photo book rather than a template.
• Whether it reflects a Hasselblad, Leica or Ricoh-style photographic temperament.
• Whether there is real composition, a sense of presence, everyday details and lens narrative.
• Whether it is clear and natural, avoiding gray haze, dirty yellow, heavy filters and over-sharpening.
• Whether it only borrows the temperament of the reference image without copying the specific image structure.
• Whether the main image is strong enough and can stand independently as a photographic work.
• Whether the auxiliary images supplement the narrative rather than supplementing a checklist of landmarks.
• Whether there is tourism-promotion color, a stock-photo feel, a postcard perspective or a small-fresh filter.
• Whether shadow weight, edge occlusion and the imperfections of a real scene are preserved.

Prompt generation rules

Write the prompt given to the image model as a photography task, not a travel guide task.

Must include:

• photobook-quality travel documentary photography collage
• one dominant cover-grade main photo
• subdued saturation, weighted shadows, soft controlled highlights
• foreground occlusion, edge crop, layered depth, off-center subject
• not a postcard, not a tourism advertisement, not a checklist of landmarks

Avoid including words that lower the quality:

• beautiful travel destination
• clear bright vibrant colors
• iconic landmark view
• perfect scenic composition
• high saturation
• clean postcard

When the user only gives a location, by default first choose "life and space observation", do not default to choosing a complete collection of the most famous sights.

Execution logic

读取用户输入
-> 识别编号条目
-> 将每个编号拆分为独立任务
-> 为当前任务分析地点、题材与相机气质
-> 规划 2-4 个不同景别和瞬间
-> 设计有主次关系的 3:4 拼图
-> 检查摄影质感、通透度与原创性
-> 单独调用图像生成
-> 按编号继续生成下一张
-> 不输出提示词和冗长说明

Output location

When the generation result needs to be saved locally, follow the user's global output directory rules, using <OUTPUT_ROOT> to indicate the user's configured output root directory:

• Images are saved by default to <OUTPUT_ROOT>\images
• Draft versions are saved to <OUTPUT_ROOT>\drafts
• Temporary files are saved to <OUTPUT_ROOT>\temp
• Final exports are saved to <OUTPUT_ROOT>\exports

Do not place newly generated large files on the system drive or in personal document directories unless the user explicitly requests it.