ChatGPT Image Generator Workflow

analyze this photo and give me a detailed JSON prompt that recreates it. break down the color grading and every exact color in the photo

(use Opus, not Sonnet. Opus has stronger visual analysis and writes more detailed JSON)

paste that JSON into ChatGPT
upload your product image and prompt:
using this JSON as reference, generate a person holding my product
save that generated photo as your character reference

attach it to every future generation for facial consistency

you now have a consistent UGC model that works across any product

the JSON controls the lighting and color grading. GPT image-2 handles the character. you control the product placement.

the #1 tell on AI photos is flat colors and a grainy look. this method removes both.
5 minutes to set up. unlimited variations after.

{"type":"Japanese YouTube thumbnail collage","style":"bold anime-tech promotional thumbnail with clean digital illustration, high contrast, saturated colors, thick outlined typography, energetic magazine-style composition","canvas":{"aspect_ratio":"16:9","resolution":"1200x630"},"background":{"frame":"mint green outer border with cyan accent strip along the top and bottom edges","main_scene":"warm modern bedroom interior with wooden headboard, beige bedding, cream curtains, bedside shelf, and a small alarm clock"},"layout":{"sections":[{"title":"データセット作り!","position":"top-left","count":8,"labels":["back view","three-quarter back view","front view","full back standing view","side profile","rear side profile","three-quarter front view","standing portrait with crossed arms"]},{"title":"部分修正!","position":"bottom-left","count":1,"labels":["character editing app screenshot with anime woman in white jacket over blue shirt"]},{"title":"素材作り!","position":"lower-mid-left","count":1,"labels":["smartphone asset design image showing a hand holding a purple screen phone with a large eye symbol and a vertical strip of 5 small variant thumbnails"]},{"title":"合成!","position":"upper-center-right","count":2,"labels":["anime woman cutout portrait","bedroom background plate"]}],"connectors":{"count":2,"style":"curved pink arrows with white outline pointing from the cutout portrait and bedroom plate toward the final composite on the right"},"text_overlays":{"count":7,"items":["データセット作り!","部分修正!","素材作り!","合成!","「{argument name=\"headline text\" default=\"GPT Image2\"}」徹底検証!","ノイズ除去法から{argument name=\"feature text\" default=\"LoRA学習\"}まで","©{argument name=\"credit name\" default=\"スタジオ真榊\"}"]}},"subjects":{"main":{"type":"anime woman","count":1,"appearance":{"age":"young adult","build":"slim curvy","hair":{"color":"{argument name=\"hair color\" default=\"long black\"}","style":"very long straight hair with soft shine, center part, strands falling over shoulders"},"face":"intentionally blurred/censored rectangle over the face","outfit":["white fitted long-sleeve scoop-neck top","visible black camisole straps","black pants"]},"pose":"seated on the edge of a bed, torso angled slightly left, hands resting near lap, looking toward camera"},"inset_character":{"type":"same anime woman reference art","count":10,"notes":"all inset figures use the same character design with face blurred in several panels"}},"typography":{"primary":"very large Japanese headline across the bottom in white with black outline and hot pink emphasis around GPT Image2","secondary":"subheadline in orange, white, black, and bright blue mixed emphasis","annotation":"handwritten energetic pink labels with white stroke near each inset section"},"color_palette":{"count":8,"colors":["hot pink","white","black","cyan","mint green","orange","beige","dark brown"]},"composition":"right half dominated by the final anime bedroom illustration, left half packed with inset process examples and screenshots, oversized bottom headline spanning nearly full width, designed to advertise a deep-dive verification article about GPT Image2 workflows for dataset creation, partial edits, asset generation, compositing, noise removal, and LoRA training"}

Ultra-clean automotive poster featuring the exact same car as the photo that will be provided later. The AI must replicate the car from the uploaded photo with identical body shape, proportions, stance, color, trims, wheels, and all visible exterior details.
The car is presented in a front three-quarter angle facing right, matching the perspective of the original reference layout, but now depicted in a subtle {argument name="action" default="drifting action"}. The drift is expressed through realistic weight transfer, slight body lift, controlled smoke plumes from the rear tires, and faint curved tire marks behind the vehicle, without distorting the original car’s geometry.
Headlights follow the exact style from the reference photo of the car, with optional warm fog lights glowing if the provided car has them. All decals, emblems, plates, and window tints must match the car from the uploaded photo.
The car drifts on a glossy white reflective studio-like floor that maintains soft reflections and realistic shadows, enhanced with light drift skid reflections and directional smudges.
Background remains a clean white-to-light-gray gradient with a giant semi-transparent bold typography of the car model name (auto-extracted from the uploaded photo) vertically dominating the background.
At the very top: clean branding text “{argument name="brand name" default="CAR COMPANY NAME"}” (or the brand detected from the uploaded car photo). Under it, spaced-out stylized tracking text containing the same brand and model name.
Below the car: centered title of the exact model name from the uploaded photo.
Under that, a short descriptive paragraph about the car’s character (efficiency, style, reliability).
Bottom section shows a clean grid layout of specifications. If real specs are known from the detected car model, generate accurate values; if not, generate placeholders in the same layout style (4 columns: horsepower, 0–100 km/h, top speed, engine displacement/fuel type).
Entire poster is minimalist, editorial, high-key studio lighting with ultra-sharp reflections, crisp shadows, modern typography, and 4K believability, blending clean design with dynamic drifting energy, Ratio 9:16

[CORE TASK]
Transform the provided input image into a pose-and-light analysis sheet.

This is NOT a finished character illustration.
This is NOT a clothing sheet.
This is NOT a beauty-preserving redraw.

This is a white-line rough mannequin conversion.

[PRIMARY GOAL]
Extract and visualize only:
- pose structure
- body balance
- camera angle
- body line flow
- inferred light source placement
- illuminated areas and light intensity

[INPUT ROLE]
Use the provided image as the strict anchor for:
- pose
- camera angle
- body tilt
- weight distribution
- approximate lighting situation

Do NOT preserve:
- face rendering
- hairstyle rendering
- clothing detail
- accessories
- weapon detail
- background architecture
- character identity
- emotional expression

[FIGURE CONVERSION]
single rough mannequin-like human figure
white body contour lines
white internal construction lines
simple mannequin head
no face
no eyes
no mouth
no eyelashes
no personality
no individual identity

human figure should look like:
- rough pose mannequin
- anatomy proxy
- line-based body guide
- structural sketch
- white-line rough dummy

keep:
- pose readability
- silhouette flow
- head tilt
- torso direction
- pelvis direction
- limb placement

[BACKGROUND]
pure black background
negative-style dark field
no scenery
no props
no architecture
no environmental storytelling

[LINE STYLE]
rough white line drawing
clean but sketch-like
construction-line feeling
anatomy guide lines visible
joint flow visible
body contour emphasized
no polished illustration finish

[LIGHT ESTIMATION]
predict the likely light source positions from the input image
visualize the light sources and illuminated areas using green glow only

use green light intensity with variation:
- strongest green where the light directly hits
- medium green for wrap light
- soft green for reflected or fading light

mark the estimated light sources with labels and arrows such as:
- Main Light
- Rim Light
- Fill Light
- Floor Bounce
- Back Light
only if appropriate

IMPORTANT:
do not invent random lights
infer lighting from the original input image
if the lighting is ambiguous, keep the annotations simple and plausible

[GREEN LIGHT VISUALIZATION]
show green glow on:
- head / skull plane
- neck
- shoulders
- chest plane
- ribcage direction
- pelvis edge
- thigh planes
- knee contact points
- floor contact bounce if applicable

use green light not as decoration,
but as lighting analysis information

[POSE PRIORITY]
1. preserve pose structure
2. preserve camera angle
3. preserve body balance
4. preserve head-torso relationship
5. visualize likely light direction
6. show illuminated areas with readable green intensity variation

[NEGATIVE]
finished person,
cute girl,
detailed face,
hair rendering,
clothing rendering,
weapon emphasis,
beautiful anatomy

Shot with a high-definition underwater camera, this prompt captures real underwater natural scenery in a {argument name="environment description" default="clear shallow stream next to a tropical primeval forest"}.

Vertical composition, 3:4 aspect ratio, medium-to-long shot. The lens is slightly below the water surface, showing the shimmering bottom of the water surface above with realistic water wave refraction and natural reflections.

Sunlight shines diagonally into the water from the top right, forming soft beams and underwater spots. Dark reflections and shadows of large tree branches occupy part of the composition in the upper right.

In the middle is clear and quiet stream water with slight suspended particles and {argument name="number of fish" default="5-8"} native freshwater small fish swimming naturally, mainly gray-silver and light brown, varying in size and distance, not forming an organized school. On the stream bed, deep green, yellow-green, and brownish-green water plants grow naturally, swaying gently with the current, distributed naturally unlike an artificial aquarium.

The bottom consists of gray-brown fine sand, gravel, pebbles, and several naturally shaped stones with slight algae marks and signs of water erosion. Multiple springs at the bottom show fine sand billowing slowly from small holes, creating light sand clouds and local water disturbances, not white smoke, steam, or large bubbles.

Natural landscape live-action photography, high-end natural documentary feel, close to National Geographic ecological photography. Features transparent water, natural lighting, restrained colors, realistic underwater optical effects, slight graininess, natural depth of field, and high-definition details.

No people, no buildings, no artificial traces, no text, no borders, no LOGO. Avoid CG feel, aquarium look, seabed coral, exaggerated fish schools, oversaturated greens, dreamy lighting effects, or plastic water plants.

Based on 1-3 clear personal photos uploaded by the user, generate a 3x3 grid photo puzzle with the theme "{argument name="photo theme" default="Summer Grape Girl Photo Series"}".

Strictly preserve the subject's real identity characteristics, including face shape, facial proportions, eye/brow structure, nose, lips, skin tone, age, hairstyle features, and overall temperament. The person in all nine images must clearly look like the same real girl; she must not become a stranger, look Westernized, look like a generic influencer, be over-beautified, or have an AI-generated face.

The overall theme is a fresh and natural everyday girl's portrait. The character wears a {argument name="clothing description" default="creamy white or off-white soft dress / slip dress"} and a {argument name="accessory" default="purple vintage floral headscarf"}. The overall look is clean, natural, and daily, with a summer girl vibe. Accessories are simple, like small earrings, but not overly ornate.

Set the scene as a summer picnic portrait in an outdoor meadow, under tree shadows, or by a vineyard. Include elements like purple grapes, grape clusters, woven baskets, glass bottles, picnic blankets, and light-colored tableware to create a natural lifestyle feel. Sunlight filters through leaves, creating soft dappled shadows, with a naturally blurred background.

Design the final image as a 3x3 grid with white borders. All nine small photos must feature the same person, same outfit, and same scene, but each must be distinctly different: different expressions, different facial angles, different poses, different camera positions, and different compositions (wide vs. close-up). Do not just have nine slight variations of the same angle.

The nine photos can show: holding grapes and smiling, lying on the grass looking at the camera, holding a grape to the mouth, organizing the basket, sitting still facing forward, a close-up of a grape against the cheek, a candid turning shot, smelling the grapes with eyes closed, and lying on the grass holding the basket. Expressions should be varied, including quiet, playful, gentle, smiling with eyes closed, naturally daydreaming, and candid laughter.

The style is Fujifilm camera texture, Japanese film photography style, realistic shooting feel, soft natural light, shallow depth of field, slight film grain, and fresh natural tones. Purple grapes should be the visual focus. The image is clean, durable, and has a sense of life and youth.

Avoid: Western faces, influencer faces, over-beautification, plastic skin, fake faces, repetitive expressions, repetitive angles, hand deformities, distorted props, messy backgrounds, studio style, illustration style, or CG feel.

Create ONE finished premium conceptual typography poster for the exact title:

"[INPUT_TEXT]"

Single poster only. No moodboard, grid, presentation board, mockup, captions, prompt text, process sheet, or sample labels.

The title "[INPUT_TEXT]" must be the dominant visual structure of the poster: huge, readable, powerful, and spelled exactly. Do not translate, shorten, replace, or misspell it. Do not add other large readable text. Optional micro catalog text is allowed only if it stays subtle and secondary.

Silently interpret the title's meaning, mood, cultural aura, symbolic associations, psychological tension, and visual rhythm. Turn that interpretation into one strong visual metaphor.

Typography is the hero. Design custom-looking letterforms whose weight, width, contrast, spacing, rhythm, distortion, negative space, edge quality, and ink texture express the temperament of the title. The type should feel intentionally designed, not like a default font.

If "[INPUT_TEXT]" refers to a widely known person, make a large editorial portrait or full / half-body figure a major visual presence, occupying roughly 40–70% of the composition. The figure should feel recognizable through aura, posture, styling, era, expression, lighting, and symbolic atmosphere, but should not copy a specific existing photograph, official poster, campaign image, logo, slogan, or copyrighted composition. The portrait must interact with the typography: overlapping the letters, emerging from them, being framed by them, casting shadows on them, breaking through them, or being partially hidden behind them.

For all other titles, use a human figure, landscape, object, or atmospheric setting only when it strengthens the meaning. It must interact with the typography and deepen the concept, not decorate it.

Use a rich but restrained 4–6 color system matched to the theme: dominant background color, primary typography color, figure / landscape tone, emotional accent color, muted support color, and subtle paper / ink texture tone. Avoid flat black-white-red defaults unless conceptually necessary.

Composition style: high-end editorial poster, museum-quality graphic design, dramatic scale, strong hierarchy, few elements, intelligent whitespace, bold flat color areas, sharp cropping, silkscreen / lithograph / risograph grain, paper fibers, subtle ink imperfections, refined visual tension.

The final image should feel like a complete visual sentence: the title, the figure or setting, the color, and the typography explain each other.

Avoid generic word art, glossy 3D lettering, random icons, stock-photo realism, cluttered collage, excessive grunge, tourist clichés, official logos, copied slogans, copied campaign aesthetics, unrelated text, and misspelled typography.

-----

INPUT_TEXT: Phoenix Rebirth

{"type":"clean Japanese business infographic poster","format":"vertical A4-like page with rounded black outer border on white background","style":"friendly flat line illustration, minimal teal and coral accent colors, soft rounded rectangles, hand-drawn business sketch feel, black Japanese gothic headline typography, thin teal dividers, small sparkle and plant icons","language":"Japanese","headline":"{argument name=\"headline text\" default=\"AIを現実のビジネスに接続する\"}","subtitle":"{argument name=\"subtitle text\" default=\"ながらAI 4トピックス\"}","layout":{"top":"large centered headline with small decorative sparkles, subtitle below between dotted horizontal lines with teal and pink dots","main_grid":"four equal rounded cards arranged in a 2 by 2 grid, connected by teal arrow icons showing flow from 1 to 2, down to 3, across to 4","bottom_banner":"rounded summary strip with a lightbulb icon at left, Japanese closing sentence in the center, and two friendly characters at right"},"sections":[{"number":"1","position":"top left","title":"{argument name=\"section 1 title\" default=\"小さく始める\"}","main_illustration":"young person in teal hoodie working on a laptop at a desk, smartphone beside them showing a cute AI robot chat interface, speech bubble with three people, small icons for basket, walking person, chat message, plant and mug","description":"身近な課題をAIで解き、MVPで反応を見る","bullet_count":3,"bullets":["身近な困りごとを見つける","小さく作って、すぐ試す","使う人の声をもとに改良する"]},{"number":"2","position":"top right","title":"{argument name=\"section 2 title\" default=\"幻想を見極める\"}","main_illustration":"person looking at a smartphone inside a pink idea cloud with car, phone, bag and bubbles; second person thinking while writing in an open notebook, wall chart with pie chart and rising bar graph, plant, mug, magnifying glass icon","description":"AI副業は魔法ではなく、事業の本質が先","bullet_count":3,"bullets":["甘い話や一時のブームに惑わされない","価値提供と収益設計を最優先する","継続できる仕組みを作る"]},{"number":"3","position":"bottom left","title":"{argument name=\"section 3 title\" default=\"制作を拡張する\"}","main_illustration":"designer seated with pen tablet facing a desktop monitor showing an image editing interface, vertical tool icons at left, small AI robot assistant in circular bubble, tablet, stylus, pen cup, network lightbulb icon","description":"AdobeとClaude連携でデザインの流れが変わる","bullet_count":3,"bullets":["AIがアイデア出しや要素生成をサポート","Photoshop・Illustratorで効率的に編集","人の創造性が、もっと加速する"]},{"number":"4","position":"bottom right","title":"{argument name=\"section 4 title\" default=\"現実に出力する\"}","main_illustration":"desktop monitor with 3D cube modeling software connected by dotted teal lines to a 3D printer printing a small cute robot figure, geometric crystal in speech bubble, finished robot mascot on the floor","description":"AIの3Dデータが、プリンターで形になる","bullet_count":3,"bullets":["デジタルのアイデアを3Dデータ化","プリンターで試作・製品化","アイデアを現実の価値に変える"]}],"bottom_summary":{"icon":"lightbulb in a pale pink circle","text":"小さく試し、現実を見極め、創造を拡張し、形にする。\nそれが、AIをビジネスにつなげる一番の近道。","characters":"two smiling young people, one in teal hoodie pointing upward and one woman in pink top, with small sparkle marks"},"visual_requirements":"Use exactly 4 numbered topic cards, exactly 3 bullet points in each card, Japanese text must be crisp and readable, maintain generous margins, rounded corners, teal card borders, coral bullet dots in sections 2 and 4, teal bullet dots in sections 1 and 3, no photorealism."}