GPT Image 2 提示词画廊

Create an original GPT Image 2 derived-play image as one 3:2 landscape scene, not a contact sheet, not a 2x3 grid, not separate cards, not a circular rules machine, not a fold-out field kit. Format: a single photographed tabletop "constraint pinball" arcade board: ramps, plungers, bumpers, drop targets, transparent lanes, scoring windows, small output printers, rule tokens, and inspection lights. It derives playable mechanics from the companion visual constraint foundry, but must look like one continuous physical game table, not six panels. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no dangerous instructions, no watermarks.

Small readable title on the front apron: "GPT-IMAGE-2 DERIVED PLAY: CONSTRAINT PINBALL".

Show six distinct playable mechanics as labeled zones on the same pinball board, each zone using different rules, inputs, and scoring evidence:
1 "Margin Multiball" - three crop balls launch through aspect-ratio rails; players score only if the tiny subject marker stays inside safe margin windows across square, vertical, and wide outputs.
2 "Mask Peek" - sliding occlusion doors cover a fictional object thumbnail; players guess and then reveal whether the generated identity stayed consistent after each blocked view.
3 "Pictogram Jackpot" - abstract symbol balls drop through shape, function, and scale bumpers to produce a tiny fictional prop print; bonus lights reward coherent object logic.
4 "Shadow Court" - a lamp-angle plunger sends a light beam across miniature objects; players challenge wrong shadow direction, missing reflection, or impossible contact points.
5 "Relation Ramp" - caption tokens for count, position, and relation travel down connected lanes; the output print is accepted only when all relation lights match the scene thumbnail.
6 "Brief Bouncer" - noisy request tokens hit safety bumpers and fiction gates; banned-risk lights stay off while a clean production brief ticket exits the board.

Visual requirements: one continuous arcade object with all six labels integrated into the playfield, visible balls/tokens in motion, score windows, accepted/rejected output tickets, and tiny generated thumbnails. Use varied accent colors on polished wood, brushed metal, glass, rubber, LEDs, paper tickets, and neutral studio background. Professional editorial product photography, crisp concept-art detail, readable labels only, playful but rigorous, not dominated by purple, beige, slate, brown, or orange.

Create an original GPT Image 2 use-case image as one 3:2 landscape scene, not a contact sheet, not a 2x3 grid, not separate cards, not a circular rules machine, not a fold-out field kit. Format: a single isometric cutaway of a fictional "visual constraint foundry" machine: conveyor belts, transparent inspection chambers, crop gates, mask tunnels, shadow rigs, symbol hoppers, proof stamps, and tiny printed output samples moving through one continuous factory-like object on a clean studio table. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no dangerous instructions, no watermarks.

Small readable title on a metal nameplate: "GPT-IMAGE-2 USE CASE FOUNDRY: CONSTRAINT ENGINES".

Show six clearly different original use-case stations integrated into the same continuous cutaway machine, each with a short readable English label on a small industrial tag:
1 "Crop Contract" - a rough source scene passes through fixed aspect-ratio gates; output prints prove subject placement, safe margins, and thumbnail readability.
2 "Occlusion Rehearsal" - simple mask plates hide and reveal parts of a fictional object; outputs test whether identity survives partial obstruction.
3 "Icon-to-Object Crosswalk" - tiny abstract pictogram tokens enter a hopper and become plausible fictional props, with comparison labels for shape, function, and scale.
4 "Light-Logic Proof" - a miniature shadow stage with one lamp angle generates before/after object renders and small check chips for shadow direction and reflection consistency.
5 "Caption-to-Scene Grounder" - plain accessibility-style scene descriptions become inspectable safe visual thumbnails, with chips for count, position, and relation accuracy.
6 "Messy Brief Sanitizer" - cluttered sticky-note requests are transformed into a clean, safe, fictional production brief and a calm output mockup; no harmful, political, sexual, or brand content.

Visual requirements: one continuous isometric machine with visible input materials, transformation channels, inspection gauges, and generated-output thumbnails for all six stations. Labels must be small but readable. Use varied accents across metal, glass, rubber belts, colored paper, and neutral tabletop; professional product-design photography with crisp concept-art precision, high detail, playful but rigorous, not dominated by purple, beige, slate, brown, or orange.

Create an original GPT Image 2 derived-play scene as one 3:2 landscape image, not a contact sheet, not a 2x3 grid, not separate poster panels. Format: a single photographed circular tabletop "rules machine" with nested rotating rings, windows, sliders, tokens, transparent overlays, a scoring rail, and tiny output prints arranged around the rings. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no watermarks.

Small readable title engraved on the machine rim: "GPT-IMAGE-2 DERIVED PLAY: LOOP ENGINE".

Derive six playable mechanics from the companion Sensorial Loops kit, but make each mechanic a different interaction rule rather than a simple visual theme. Show labels as small engraved tabs on the same circular machine, not as separate panels:
1 "Aroma Semaphore" - players place scent-note tiles into color/texture windows, then guess the hidden room mood from only material swatches.
2 "Waveform Sculptor" - sound strips pass through silhouette gates; players score objects by how well rhythm, blur, and shadow stay consistent.
3 "Palace Breadcrumbs" - memory-route tokens must connect fact icons through a miniature room maze without repeating an anchor.
4 "Grip Auction" - players bid constraint chips on fictional handle prototypes: reach, twist, pinch, and balance; highest comfort score wins.
5 "Truth Ring" - material cards rotate under an impossible-detail detector; players flag fake seams, wrong reflections, and weight clues.
6 "Errand Reality Lock" - surreal errand cards must be grounded by time, location, object, and safety tabs before an output print is accepted.

Visual requirements: one continuous mechanical object, visible rules components and score evidence for all six mechanics, no rectangular six-panel layout. Premium tabletop prototype photography mixed with crisp concept-art output thumbnails, varied accent colors on brushed metal, paper, translucent plastic, and fabric, readable labels only, professional editorial lighting, high detail, playful but rigorous.

Create an original GPT Image 2 use-case scene as one 3:2 landscape image, not a contact sheet, not a 2x3 grid, not separate poster panels. Format: a single overhead photographed fold-out "visual instrument case" on a neutral workbench, like a compact field kit opened flat with drawers, transparent sleeves, rotating dials, reusable masks, acetate overlays, and tiny printed output samples. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no watermarks.

Overall small title stamped on the case: "GPT-IMAGE-2 USE CASE KIT: SENSORIAL LOOPS".

Show six clearly different original use-case stations integrated into the same physical kit, each with a short readable English label:
1 "Scent Palette Translator" - fictional aroma note cards become color, texture, and lighting studies for a room scene; show non-edible scent tokens and fabric swatches, no real products.
2 "Sound Shape Notation" - abstract sound-wave strips become geometric object silhouettes, motion trails, and lighting rhythm studies.
3 "Memory Palace Map" - a plain list of fictional study facts becomes a walkable miniature room route with icon anchors and recall checkpoints, no medical claims.
4 "Gesture-to-Tool Mockup" - hand-motion arrows and simple ergonomic constraints become fictional tool handle shape prototypes, with comfort score stickers.
5 "Material Honesty Audit" - the same object is rendered as ceramic, paper pulp, glass, rubber, and stitched felt with small failure tags for impossible seams/reflections.
6 "Dream Errand Planner" - surreal but safe errand cards become practical storyboard thumbnails for a fictional day plan, with reality-check tabs.

Visual requirements: all six labels are placed on small tabs within the case, but the composition remains one continuous object with drawers and overlays, not six rectangular panels. Include input materials, generated-output thumbnails, and evaluation tokens for each station. Premium editorial product photography, crisp readable typography for labels only, varied accent colors on neutral materials, professional lighting, high detail, no one-note purple or beige palette.

Create a second polished original GPT Image 2 derived-play mechanics board as a single 3:2 landscape image, clean 2x3 contact-sheet layout with six clearly separated panels. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no watermarks. Overall tiny header: "GPT-IMAGE-2 DERIVED PLAY MODES: MATERIAL ARCADE". Derive playful, recordable image-generation games from the prior material/accessibility use cases. Each panel has a short readable English label and looks like a complete tabletop or studio challenge card:
1 "Texture Cipher Relay" - players convert everyday object photos into raised-line texture keys, then another player reconstructs the object category from the diagram.
2 "Drawer Tetris Draft" - randomized fictional tools must fit into a foam insert grid; show tokens, grid slots, and scored layout cards.
3 "Weather Oracle Cards" - material sample cards are dealt with sun/rain/frost/dust modifiers to predict visible aging outcomes.
4 "Prototype Punch-Out Duel" - two paper-kit designs compete through constraints: fewest folds, strongest shape, smallest scrap pile.
5 "Mood Mascot Mixer" - emotion words, color chips, and shape dice combine into non-human mascot families with consistency rules.
6 "Mini-Tour Stamp Rally" - fictional exhibit objects become route stamps; players build the clearest three-stop visitor path from shuffled object cards.
Style: collectible creative game-design spread, crisp typography, realistic cards and tokens photographed on a clean studio table, thin dividers, restrained varied accent colors, legible labels, professional editorial design, no brand marks.

Create a polished original GPT Image 2 use-case research board as a single 3:2 landscape image, clean 2x3 contact-sheet layout with six clearly separated panels. No logos, no real brands, no celebrity likenesses, no public figures, no copyrighted characters, no politics, no sexual content, no violence, no watermarks. Overall tiny header: "GPT-IMAGE-2 USE CASE LAB: MATERIAL + ACCESSIBILITY PLAY". Each panel has a short readable English label and shows a fictional, practical image-generation workflow:
1 "Tactile Diagram Draft" - a fictional kitchen timer and plant pot transformed into high-contrast raised-line style diagrams with braille-like dotted texture placeholders, accessibility education concept.
2 "Drawer Shadow Planner" - a messy set of fictional craft tools converted into a neat foam-insert drawer layout with silhouettes and tiny labels.
3 "Season Wear Simulator" - one generic outdoor stool shown across sun, rain, frost, and dust aging previews, material planning concept.
4 "Cardboard Kit Forge" - a rough sketch of a small imaginary desk organizer turned into punch-out cardboard parts and assembled prototype steps.
5 "Mascot Mood Notes" - abstract daily moods converted into friendly non-human geometric mascots on journaling cards, no therapy claims.
6 "Fictional Exhibit Mini-Tours" - a invented object collection turned into small visitor route thumbnails and plain-language captions, all fictional.
Style: premium design lab documentation wall, crisp typography, realistic paper textures, neutral white background with restrained accent colors, accurate readable labels, professional editorial lighting, no brand marks.

Create a second polished GPT Image 2 derived-play board as a single 3:2 landscape image, laid out as a clean 2x3 contact sheet with six distinct panels. This board should derive playful mechanics from the previous use cases, with compact legible English panel labels. No giant title, no logos, no watermarks.

Panel 1 label: AVATAR CONSENT DICE. Show a tabletop-style rules mat with dice for pose, camera angle, outfit coverage, and approval states; the game is about generating only consent-safe avatar variants.
Panel 2 label: ASSEMBLY TIME TRIAL. Show three lamp assembly snapshots as a timed puzzle lane, with ghost overlays, error tokens, and a final alignment score.
Panel 3 label: CURATOR SWITCHBOARD. Show one artifact at the center and toggle cards for audience age, reading level, language tone, and label length; outputs become different museum cards.
Panel 4 label: FIXTURE BOSS RUSH. Show a QA arcade board with synthetic image challenges: glare, blur, occlusion, tiny text, low light, reflection; each has a multimodal test badge.
Panel 5 label: ROOM MOOD SPELLBOOK. Show a recipe-like visual spellbook for transforming a room mood with lighting, object swaps, texture cards, and forbidden style clashes.
Panel 6 label: RULEBOOK MUTATION LAB. Show a controlled mutation lab for a fictional game item: sliders for silhouette, palette, scale, background noise, and icon readability, with approved/rejected outputs.

Style: refined visual design, research-lab meets game manual, tactile paper and screen mockups, warm gray background with varied accent colors, clear panel boundaries, professional composition, no decorative blobs, avoid a one-note purple/blue palette. Keep all content fictional and privacy-safe.

Create a polished GPT Image 2 use-case research board as a single 3:2 landscape image, laid out as a clean 2x3 contact sheet with six distinct panels. No giant title, no logos, no watermarks. Each panel should show a different novel use case concept with small legible panel labels in English only.

Panel 1 label: CONSENT AVATAR WARDROBE. Show a fictional person-safe avatar mannequin, outfit layers, camera-angle lock icons, and a style approval checklist; communicate identity-preserving fashion/character exploration without using a real celebrity.
Panel 2 label: AR ASSEMBLY GHOSTS. Show a desk with a flat-pack lamp, translucent step overlays, screw callouts, and before/after component alignment; a use case for turning manuals into visual assembly guidance.
Panel 3 label: MUSEUM LABEL REMIX. Show one artifact rendered in three exhibit label styles: kid-friendly, expert catalog, and audio-guide card; emphasize controlled text + visual context.
Panel 4 label: VISION TEST FIXTURES. Show small synthetic screenshots, bounding boxes, edge cases, and pass/fail stickers; a use case for generating QA fixtures for multimodal app testing.
Panel 5 label: MOOD TO ROOM SET. Show the same small room transformed into calm focus, celebratory dinner, and rainy reading modes using lighting, objects, and color.
Panel 6 label: IMAGE RULEBOOK. Show a visual rule sheet for a fictional game item, with icon constraints, forbidden variants, and example cards.

Style: sophisticated editorial product-research board, realistic mixed-media mockups, crisp small text where possible, neutral background, strong visual separation between panels, no purple/blue gradient theme, no decorative blobs. Use consistent margins and stable panel framing.

Create a second polished GPT-image-2 derived-play research board as one single 3:2 landscape image, like a collectible creative game-design spread. Overall title: "GPT-IMAGE-2 DERIVED PLAY MODES: CONTEXT LOOP ARCADE". Six clearly separated panels in a clean 3x2 grid. These play mechanics are derived from the companion use-case board "Live Context Loops" and must be NEW, avoiding all prior recorded gallery mechanics including prompt escape room, brand multiverse, time machine, recipe comic, worldbuilding atlas, prompt archaeology, QA spot-the-diff, microbrand kit, learning card deck, wayfinding remix, tabletop quest props, state lock, evidence garden, scenario switch, prep race, palette detective, failure ladder, observation bingo, constraint auction, recipe forensics, policy storyboard race, archive detective, glitch taxonomy draft, and any direct repeats of calendar scene forge, meeting memory board, dataset postcard sampler, repair triage photos, language mood matrix, spec-to-shelf prototype.

Panel 1 title: "Week Remix Draft" — players draw event cards, weather chips, and outfit tokens, then score whether seven generated day-scenes preserve time, place, and continuity.
Panel 2 title: "Decision Thread" — teams reconstruct a meeting outcome from visual action cards, blocker stamps, owner chips, and a hidden missing-decision token.
Panel 3 title: "Fact Postcard Duel" — players turn spreadsheet rows into travel postcards, then opponents audit every visible number, unit, and category icon for fidelity.
Panel 4 title: "Triage Relay" — appliance issue photos become numbered checklist boards; players race to place severity markers without giving unsafe repair advice.
Panel 5 title: "Tone Switchboard" — one message is routed through mood sliders and script tiles; players score if visual tone changes while the core meaning stays intact.
Panel 6 title: "Shelf Spec Poker" — product requirement cards are bet into packaging mockups; winning hands balance warnings, claims, ingredients, insert steps, and shelf contrast.

Style: premium tabletop prototype photography plus polished concept art, tactile cards, modular tiles, sliders, score sheets, chips, rulers, pencils, small generated-output thumbnails. Each panel should show components, scoring evidence, and a tiny play log. Use varied accent colors on a neutral table, crisp readable typography only for the overall title, six panel titles, and simple labels like "SCORE", "CHECK", "SAFE", "FACT", "MOOD", "CLAIM". No real brands, no logos, no watermarks, no copyrighted characters, no celebrity likenesses.

Create a polished GPT-image-2 use-case research board as one single 3:2 landscape image, like a premium design lab documentation wall. Overall title: "GPT-IMAGE-2 USE CASE LAB: LIVE CONTEXT LOOPS". Six clearly separated panels in a clean 3x2 grid, each panel with a short readable English title and a concrete demo. These must be NEW use cases, avoiding prior recorded gallery themes: multilingual posters, product SKU factories, UI fiction simulators, storyboard continuity, educational remixes, interior matrices, prompt escape rooms, brand multiverses, time machines, recipe comics, worldbuilding atlases, catalog transformers, data posters, onboarding state machines, localized packaging, restoration what-ifs, game asset sheets, prompt archaeology, QA spot-the-diff, microbrand kits, learning cards, wayfinding remix, tabletop quest props, process twins, field notebooks, policy theater, kitchen mise maps, archive color labs, failure mode atlases, state lock, evidence garden, scenario switch, prep race, palette detective, failure ladder, field note builder, constraint posterizer, recipe reality check, policy comic translator, archive label restorer, failure mode catalog, observation bingo, constraint auction, recipe forensics, policy storyboard race, archive detective, glitch taxonomy draft.

Panel 1 title: "Calendar Scene Forge" showing a week of plain calendar events transformed into consistent location thumbnails, outfit hints, weather badges, and time-of-day lighting cues.
Panel 2 title: "Meeting Memory Board" showing rough meeting notes converted into a visual decision map with faces as neutral silhouettes, action cards, blockers, and ownership chips.
Panel 3 title: "Dataset Postcard Sampler" showing rows from a small fictional spreadsheet turned into six mini postcards, each preserving numeric facts as readable labels and icons.
Panel 4 title: "Repair Triage Photos" showing user-submitted appliance photos becoming a safe diagnostic photo checklist with numbered areas, severity tags, and "call expert" flags.
Panel 5 title: "Language Mood Matrix" showing the same short message rendered as visual tone cards across formal, friendly, urgent, playful, and calm moods in multiple scripts, no real brands.
Panel 6 title: "Spec-to-Shelf Prototype" showing a plain product requirements spec converted into shelf-ready rough mockups: box, label, insert, warning card, and comparison tag.

Style: premium editorial documentation, realistic printed boards, crisp cards, mini screenshots, tiny charts, material samples, sticky notes, calipers, pencil annotations, warm studio light. Rich varied accent colors but not dominated by one hue. Typography should be readable only for overall title, six panel titles, and short labels like "TODO", "RISK", "OWNER", "FACT", "SAFE", "DRAFT". No real logos, no watermarks, no copyrighted characters, no celebrity likenesses.

Create a second polished GPT-image-2 derived-play research board as a single 3:2 landscape image, like a collectible creative game design spread. Six clearly separated panels in a 3x2 grid, each panel with a short readable English title and a distinct playable demo. Overall title: "GPT-Image-2 Derived Play Modes: Evidence to Systems".

These must be NEW play mechanics derived from the companion board: Field Note Builder, Constraint Posterizer, Recipe Reality Check, Policy Comic Translator, Archive Label Restorer, and Failure Mode Catalog. Avoid repeating earlier recorded mechanics such as prompt archaeology, QA spot-the-diff, microbrand kit, learning card deck, wayfinding remix, tabletop quest props, rubric arena, lighting DJ, fix-it quest, signage escape room, manual maze, prop continuity poker, prompt relay, constraint bingo, gallery speedrun, material truth table, or source stamp sprint.

Panel 1 title: "Observation Bingo" showing players marking generated field-note cards for date, location, subject, evidence, uncertainty, and sketch clarity.
Panel 2 title: "Constraint Auction" showing poster briefs as auction lots, with bid tokens for ink limit, contrast rule, object count, and copy length.
Panel 3 title: "Recipe Forensics" showing plated food image cards matched against ingredient clues, timing tracks, allergen tokens, and missing-step alerts.
Panel 4 title: "Policy Storyboard Race" showing teams turning rule cards into four-frame comic strips, judged by sequence clarity and misuse prevention.
Panel 5 title: "Archive Detective" showing museum label fragments, accession-number tiles, object-photo cards, era stamps, and reconstruction score sheets.
Panel 6 title: "Glitch Taxonomy Draft" showing players sorting generated-image failure cards into categories: text warp, scale drift, shadow gap, repeat object, and fix candidate.

Style: premium tabletop design-game photography plus polished concept art, tactile cards, tokens, score tracks, magnifier tools, pencils, and small generated-output thumbnails. Rich but restrained color variety, warm studio documentation lighting, crisp typography only for the overall title, six panel titles, and simple game labels. No real brands, no logos, no watermark, no copyrighted characters. Make every panel visually distinct, recordable, and immediately understandable as a game people could play.

Use case: stylized-concept
Asset type: GPT Image 2 derived play modes board, 3:2 landscape contact sheet
Primary request: Derive six NEW playful, recordable experiments from the "System Stress Tests" use-case board. Do not repeat earlier gallery mechanics such as prompt relay, constraint bingo, regression detective, variant draft arena, scene grammar kit, scale-shift microscope, wardrobe audit, gallery speedrun, drill-path puzzle, material truth table, waveform jam, maintenance triage, claim court, access route remix, director's cut ladder, aging oracle, taste telemetry duel, edge-case hunt, manual maze, price canon kitchen, prop continuity poker, signage babel sprint, figure ladder duel, locked-room remix, rubric arena, lighting DJ, fix-it quest, source stamp, alpha edge, ratio relay, safety sketch, tiny type, deck poker, material guessing, plant doctor, clause boss battle, bookmark dungeon, legibility gauntlet, aging timeline poker, prompt archaeology, QA spot-the-diff, microbrand kit, learning card deck, wayfinding remix, or tabletop quest props.
Overall title: "GPT-IMAGE-2 DERIVED PLAY MODES: OPERATIONS ARCADE".
Panel 1 title: "State Lock" — players generate four process states and score whether the same machine remains structurally identical across time.
Panel 2 title: "Evidence Garden" — teams convert rough nature notes into observation plates, then other players guess which visual claims are certain, partial, or speculative.
Panel 3 title: "Scenario Switch" — policy scenes are remixed by swapping one variable at a time: tone, setting, power dynamic, and outcome, while preserving the rule.
Panel 4 title: "Prep Race" — recipe station cards become a timed sequencing game where players assemble mise en place before a clock token reaches service.
Panel 5 title: "Palette Detective" — players match recolored archive objects to evidence swatches and justify which palette is most plausible.
Panel 6 title: "Failure Ladder" — device condition cards are arranged from normal to severe, with repair tokens and risk markers scored for continuity.
Style/medium: premium tabletop game documentation spread, realistic cards, score mats, tokens, timers, loupes, pencils, small generated-output samples, no real brands.
Composition/framing: six equal panels in a clean 3x2 grid; each panel shows title, components, scoring evidence, and a tiny play record sheet.
Lighting/mood: warm studio tabletop lighting, playful but practical, immediately understandable.
Color palette: neutral table, varied accent colors for each game, crisp black rule text, colorful scoring tokens.
Text constraints: render only the overall title, six panel titles, and simple words "PASS", "RISK", "FIX", "SCORE", "ROUND"; no logos, no watermarks, no copyrighted characters.

Use case: productivity-visual
Asset type: GPT Image 2 use-case research board, 3:2 landscape contact sheet
Primary request: Create a polished research board titled "GPT-IMAGE-2 USE CASE LAB: SYSTEM STRESS TESTS" showing six NEW practical use cases for image generation and editing that avoid prior themes such as catalog transformation, data posters, onboarding states, localization packaging, restoration, game asset sheets, prompt archaeology, QA spot-the-diff, microbrand kits, learning cards, wayfinding, tabletop props, transparent stickers, aspect ratio relays, safety sketches, menus, style decks, repair manuals, room lighting, legal risk maps, bookmark memory palaces, event poster stress labs, and object aging.
Panel 1 title: "Process Twin" — a factory maintenance workflow visualized as before, during, after, and exception states with the same machine silhouette preserved.
Panel 2 title: "Field Notebook" — messy field notes and specimen sketches transformed into a clean ecological observation plate with readable labels and uncertainty badges.
Panel 3 title: "Policy Theater" — abstract workplace policy paragraphs turned into role-play scene cards showing safe and unsafe examples, no real companies.
Panel 4 title: "Kitchen Mise Map" — a recipe converted into station setup diagrams, timing strips, ingredient prep bowls, and plating thumbnails.
Panel 5 title: "Archive Color Lab" — black-and-white public-domain-style object photos recolored into three historically plausible palettes with evidence swatches.
Panel 6 title: "Failure Mode Atlas" — one everyday device shown across jammed, overheated, cracked, water-exposed, and repaired states with callout tags.
Composition/framing: six equal panels in a clean 3x2 grid, each with a short readable English title, visual evidence artifacts, and a finished sample output.
Style/medium: premium design-research documentation wall, crisp printed boards, realistic desk objects, high detail, professional lighting.
Color: varied accents per panel, neutral workspace base, avoid a one-hue palette.
Text constraints: render only the title plus the six panel titles and tiny simple labels like "OK", "RISK", "FIX", "SOURCE"; no real brands, no logos, no watermark, no political figures, no copyrighted characters.

Create a polished GPT-image-2 use-case research board as a single 3:2 landscape image, like a high-end design lab documentation artifact. Six clearly separated panels in a 3x2 grid, each panel with a short readable English title and a distinct visual demo. Overall title: "GPT-Image-2 Use Case Lab: Evidence to Systems".

Panel 1 title: "Field Note Builder" showing messy handwritten field notes transformed into a clean illustrated observation sheet with icons, map pins, specimen thumbnails, and a dated log strip.
Panel 2 title: "Constraint Posterizer" showing one plain prompt becoming three poster variants under visible constraints: low ink, high contrast, and one-object-only.
Panel 3 title: "Recipe Reality Check" showing a fictional recipe card converted into plated food, ingredient layout, timing diagram, and allergy badge system.
Panel 4 title: "Policy Comic Translator" showing a dense workplace policy paragraph transformed into a four-panel comic explainer with clear visual steps and readable labels.
Panel 5 title: "Archive Label Restorer" showing damaged museum object labels reconstructed into plausible clean catalog cards with accession codes, era tags, and object thumbnails.
Panel 6 title: "Failure Mode Catalog" showing a generated product image audit wall: warped text, missing shadow, wrong scale, repeated object, and corrected examples.

Style: premium editorial collage, crisp realistic miniatures mixed with clean interface-like labels, varied colors across panels, professional lighting, sharp details.
Text constraints: render only the overall title and the six panel titles exactly as listed plus tiny simple labels where needed; avoid long unreadable paragraphs; no logos, no brands, no watermark.
Composition: six equal panels in a clean 3x2 grid, balanced margins, print-quality contact sheet.

Create a second polished GPT-image-2 derived-play research board as a single 3:2 landscape image, like a collectible playbook spread for creative workflows. Six clearly separated panels in a 3x2 grid, each panel with a short readable English title and a distinct visual demo. Overall title: "GPT-Image-2 Derived Play Modes: Batch 2026-04-25 B". Panel 1 title: "Prompt Archaeology" showing a final image being reverse-engineered into visual clues, prompt cards, and a timeline of iterations. Panel 2 title: "QA Spot-the-Diff" showing generated ad variants with redline annotations for text errors, object drift, and layout mismatch. Panel 3 title: "Microbrand Kit" showing a fictional street tea brand expanded into logo, cup, poster, menu board, and social tile. Panel 4 title: "Learning Card Deck" showing a science concept turned into a set of illustrated flashcards with consistent icons and difficulty levels. Panel 5 title: "Wayfinding Remix" showing airport signage localized into multiple languages and accessibility modes while preserving geometry. Panel 6 title: "Tabletop Quest Props" showing a map, item cards, NPC portrait, puzzle sheet, and reward token in a coherent campaign style. Crisp typography, no real brands, no watermarks, rich but restrained color, sharp details, professional gallery-ready composition.

Create a polished GPT-image-2 use-case research board as a single 3:2 landscape image, like a high-end editorial design research artifact. Six clearly separated panels in a 3x2 grid, each panel with a short readable English title and a distinct visual demo. Overall title: "GPT-Image-2 Use Case Lab: Batch 2026-04-25 A". Panel 1 title: "Catalog Transformer" showing one product silhouette becoming four catalog-ready product photos with different materials. Panel 2 title: "Data-to-Poster" showing a small chart and numbers transformed into a clean infographic poster. Panel 3 title: "Onboarding States" showing mobile app screens for empty, loading, success, and error states. Panel 4 title: "Localized Shelf" showing the same fictional snack packaging localized for English, Japanese, and Arabic with accurate-looking layout diversity, no real brands. Panel 5 title: "Restoration What-if" showing an old damaged storefront photo reconstructed into plausible restored facades across eras. Panel 6 title: "Game Asset Sheet" showing a consistent fantasy tool kit with icons, item cards, and a small inventory UI. Crisp typography, no logos, no watermarks, rich but restrained color, sharp details, professional gallery-ready composition.

Use case: stylized-concept
Asset type: 3x2 game-mode board for a GPT Image 2 prompt gallery
Primary request: Derive six playful game mechanics from the practical use cases in the previous board. Make each panel look like a complete tabletop or party-game prototype that people could actually play with generated images.
Panel titles and scenes:
1. Material Guessing League — players inspect object variants and place tokens guessing material, cost, weight, and durability without seeing the prompt.
2. Plant Doctor Speedrun — teams diagnose generated plant evidence cards before a timer runs out, matching symptom tiles to care-plan tiles.
3. Clause Boss Battle — scary contract phrases become monsters; players defeat them with negotiation, safeguard, and plain-language rewrite cards.
4. Bookmark Dungeon Crawl — a bookmark list becomes a tiny dungeon map; players navigate rooms by reconstructing the original categories from visual clues.
5. Legibility Gauntlet — the same event title appears on shrinking media pieces; players score text readability, hierarchy, and brand-free consistency.
6. Aging Timeline Poker — players arrange generated object-aging cards in the right sequence and bet on which details stayed consistent.
Style: tactile board-game photography plus polished concept art, colorful tokens, dice, cards, miniatures, score tracks, warm table lighting, varied materials, crisp typography only for the six panel titles and simple game labels.
Composition: six equal panels in a clean 3x2 grid, no nested cards beyond game components, no logos, no brands, no watermark, no visible instructions paragraph.
Quality: high detail, playful but professional, balanced color variety, each game visually distinct and immediately understandable.

Use case: productivity-visual
Asset type: 3x2 visual playbook contact sheet for a GPT Image 2 prompt gallery
Primary request: Create six distinct, newly invented GPT Image 2 use-case cards that demonstrate high-fidelity input handling, precise text/layout rendering, controlled variation, and multi-step visual thinking. Each panel should feel like a polished example image, not an instruction page.
Panel titles and scenes:
1. Texture Passport Lab — the same simple chair silhouette rendered as walnut, translucent resin, brushed steel, quilted fabric, cork, and recycled plastic, arranged like a material sample passport with tiny swatches.
2. Plant Clinic Evidence Board — a houseplant diagnosis board with macro leaf closeups, symptom pins, watering timeline, care icons, and a calm before/after recovery preview.
3. Tiny Legal Risk Map — a playful visual audit of a short contract clause, with highlighted phrases becoming hazard signs, shields, negotiation tokens, and plain-language side notes.
4. Memory Palace From Bookmarks — a browser bookmark list transformed into a navigable miniature library room where shelves, doors, and desk objects represent saved links.
5. Event Poster Stress Lab — one short event title tested across poster, ticket, wristband, projection screen, and phone lock screen, checking legibility and layout consistency.
6. Object Aging Forecast — one everyday backpack shown new, one year used, repaired, rain-soaked, sun-faded, and archived, preserving shape and brand-free details.
Style: premium editorial collage, crisp realistic miniatures mixed with clean interface-like labels, varied colors across panels, no single-hue palette, high detail, professional lighting.
Text constraints: render only the six panel titles exactly as listed; avoid tiny unreadable paragraphs; no logos, no brands, no watermark.
Composition: six equal panels in a clean 3x2 grid, each panel visually different, balanced margins, print-quality contact sheet.

Use case: stylized-concept
Asset type: GPT Image 2 gallery derived play modes board, 2K landscape contact sheet
Primary request: Create a second polished 3x2 contact sheet of six NEW game-like play mechanics derived from the companion Images 2.0 use-case board. These must be playable frameworks for experimenting with GPT Image 2 outputs, not repeats of earlier recorded mechanics such as product-to-flavor-lab, app-to-board-game, poster-to-mini-comic, Claim Court, Access Route Remix, Director's Cut Ladder, Aging Oracle, Taste Telemetry Duel, Edge-Case Hunt, Manual Maze, Price Canon Kitchen, Prop Continuity Poker, Signage Babel Sprint, Figure Ladder Duel, Locked-Room Remix, Rubric Arena, Lighting DJ, Fix-It Quest, prompt relay, constraint bingo, regression detective, variant draft arena, scene grammar kit, scale-shift microscope, wardrobe audit, gallery speedrun, drill-path puzzle, material truth table, waveform jam, or maintenance triage.
Panels:
1. Source Stamp Sprint: players draft a current-data visual card, then earn points for visible date, source, uncertainty, and non-misleading summary cues.
2. Alpha Edge Trial: players generate cutout sticker sheets, then inspect corners, holes, shadows, and hairline edges with loupe tokens and pass/fail mats.
3. Ratio Relay Run: teams adapt one image brief across 3:1, 1:1, 9:16, and poster formats; penalties for changed hero object, lost message, or cramped text.
4. Safety Sketch Court: players convert rough product sketches into safe mockups, defending each change with hazard cards and material tokens.
5. Tiny-Type Tavern: teams build dense multilingual menu boards; judges order items using only the generated signs, scoring legibility and hierarchy.
6. Deck Pick Poker: each player generates a style deck from one brief, then bids on which direction best balances clarity, cost, novelty, and reuse.
Style/medium: premium tabletop game documentation spread, realistic cards, score sheets, timer tiles, loupes, tokens, mini output images, pencils, no real brands.
Composition/framing: 3 columns by 2 rows, six clearly separated panels; each panel shows title, components, round structure, scoring evidence, and a small generated-output sample.
Lighting/mood: warm studio tabletop lighting, collaborative design lab energy, precise enough for documentation.
Color palette: neutral table, varied accent colors for each game, crisp black rule text, colorful scoring tokens.
Text (verbatim, keep titles simple and legible): "SOURCE STAMP", "ALPHA EDGE", "RATIO RELAY", "SAFETY SKETCH", "TINY TYPE", "DECK POKER", "PASS", "FIX", "SCORE".
Constraints: no logos, no watermarks, no copyrighted characters, no real news claims, no political figures; every play mode should feel playable, scoreable, and recordable.

Use case: stylized-concept
Asset type: GPT Image 2 gallery research board, 2K landscape contact sheet
Primary request: Create a polished 3x2 visual research board showing six NEW practical and playful GPT Image 2 use cases based on current Images 2.0 capabilities: stronger instruction following, dense in-image text, high-fidelity image inputs, flexible aspect ratios, transparent/detail-oriented image edits, multi-output generation, and thinking-mode planning. Avoid repeating prior gallery themes such as repair manuals, museum wayfinding, continuity prop bibles, visual QA rubrics, same-room lighting, receipt-to-menu worlds, research figure remixers, interior swaps, policy ad labs, accessibility simulators, script blocking, future wear forecasts, flavor atlases, synthetic dataset gap fillers, board-game conversions, prompt relay, constraint bingo, regression detective, claim court, director ladder, and signage escape rooms.
Panels:
1. Live Fact Card Forge: a research desk where a model plans a fresh web-informed infographic card, with date stamps, source tabs, thumbnail map, and a final concise explainer card.
2. Transparent Sticker Factory: a creator table making clean cutout sticker sheets from ordinary objects, with alpha checkerboard previews, edge inspection loupe, and export tiles.
3. Aspect Ratio Shape-Shifter: one campaign concept shown as horizontal banner, square post, vertical story, and ultra-tall poster while preserving the same character, prop, and message.
4. Sketch-to-Safety Prototype: a rough napkin sketch of a household gadget becomes three safe product mockups with rounded edges, warning icons, and material callouts.
5. Multilingual Tiny-Type Menu: a compact fictional cafe board with readable English, Chinese, Japanese, Korean, and Arabic labels arranged in dense but clean typography.
6. Multi-Output Style Deck: eight coordinated visual directions generated from one brief, displayed as a selector board with style chips, risk notes, and pick markers.
Style/medium: premium design-research wall, realistic printed boards, screens, cutout previews, prompt cards, thumbnails, annotation stickers, no real brands.
Composition/framing: 3 columns by 2 rows, each panel visually distinct with a short simple title; visible before-to-after or source-to-output relationship in every panel.
Lighting/mood: bright studio documentation lighting, crisp and inventive, production-ready rather than whimsical.
Color palette: neutral workspace, black ink, teal/amber/red/blue accent systems, varied per panel.
Text (verbatim, keep simple and legible where possible): "FACT CARD", "TRANSPARENT", "WIDE", "SQUARE", "VERTICAL", "SAFE MOCKUP", "MENU", "STYLE DECK".
Constraints: no logos, no watermarks, no real company marks, no copyrighted characters, no real news claims, no political figures; make every panel feel like a recordable use case with components and evidence.

Use case: productivity-visual
Asset type: derived play concept image, single finished image
Primary request: Create a 16:9 concept called "RUBRIC ARENA", derived from the visual QA rubric. It turns image evaluation into a playful tournament bracket where generated image variants compete on criteria like text accuracy, layout match, and object count.
Scene/backdrop: clean judge table with bracket board, score tokens, variant cards, and rubric dice.
Subject: eight image variant cards in a tournament bracket, criteria badges, final winner pedestal.
Style/medium: polished tabletop strategy game prototype with crisp printed components.
Composition/framing: overhead view, bracket path clearly visible.
Lighting/mood: bright, analytical, fun but not childish.
Color palette: white board, charcoal text, green pass tokens, amber warning tokens, red retry tokens.
Text (verbatim): "RUBRIC ARENA", "Text", "Layout", "Objects", "Safety", "PASS", "FIX", "WINNER".
Constraints: no real brands, no logos, no watermark, avoid copyrighted game references, text should be legible.

Use case: lighting-weather
Asset type: derived play concept image, single finished image
Primary request: Create a 16:9 interactive concept called "LIGHTING DJ", derived from the same-room four-hours image. Show a tactile console that lets players remix one room's lighting with sliders for sun angle, lamp warmth, window weather, and shadow softness, while preview thumbnails update around it.
Scene/backdrop: design studio desk with a physical control board and four preview screens of the same reading nook.
Subject: knobs and sliders labeled for lighting variables, thumbnails showing consistent room with different light recipes.
Style/medium: cinematic product photography mixed with clear UI screens.
Composition/framing: console centered, preview screens behind, hands absent.
Lighting/mood: moody studio with colorful but restrained light accents.
Color palette: charcoal console, white labels, amber lamp glow, cool blue weather preview, green plant accents.
Text (verbatim): "LIGHTING DJ", "Sun angle", "Lamp warmth", "Rain", "Shadow soft", "Preview".
Constraints: no real brands, no logos, no watermark, keep the same room layout in all thumbnails, labels should be readable.

Use case: infographic-diagram
Asset type: derived play concept image, single finished image
Primary request: Create a 16:9 tabletop puzzle concept called "FIX-IT QUEST DECK", derived from the repair manual maker. It turns an exploded product diagram into a cooperative card game: players collect part cards, tool tokens, and sequence cards to assemble a lamp.
Scene/backdrop: tabletop with a small lamp model, exploded-view board, part cards, tool tokens, and a progress track.
Subject: cards labeled with lamp parts and simple icons, not actual repair instructions.
Style/medium: polished board game prototype photography with clean technical illustration on cards.
Composition/framing: overhead angle, organized play area, clear card hierarchy.
Lighting/mood: warm maker-space lighting, playful but practical.
Color palette: cream cards, graphite ink, orange highlights, steel gray pieces.
Text (verbatim): "FIX-IT QUEST", "SHADE", "HINGE", "SPRING", "BASE", "SWITCH", "ASSEMBLE".
Constraints: fictional game, no real brands, no watermark, avoid unsafe wiring steps, keep text legible.

Use case: stylized-concept
Asset type: derived play concept image, single finished image
Primary request: Create a 16:9 game-like concept called "SIGNAGE ESCAPE ROOM", derived from the museum wayfinding idea. Players solve a path through a tiny isometric museum by reading multilingual signs, matching arrows, and finding the correct exit.
Scene/backdrop: miniature isometric museum maze on a tabletop, with room tiles, signposts, colored route tokens, and clue cards.
Subject: playable puzzle board with clear arrows and a marked start/end route.
Style/medium: high-end board game product photography plus crisp miniatures.
Composition/framing: top-down three-quarter view, all pieces visible, tactile and playable.
Lighting/mood: bright tabletop game night, clever and inviting.
Color palette: teal route paths, amber clue cards, white museum walls, black text.
Text (verbatim): "START", "入口", "LIGHT", "CITY", "EXIT", "解锁".
Constraints: no real brands, no logos, no watermark, avoid copyrighted board-game references, text should be legible.

Use case: ui-mockup
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a 16:9 high-fidelity software mockup called "VISUAL QA RUBRIC" where a designer compares generated image variants against a checklist: text accuracy, object count, layout match, lighting, and safety. This demonstrates using GPT Image 2 outputs as testable artifacts.
Scene/backdrop: desktop web app interface, not a marketing landing page.
Subject: three image thumbnails on the left, scoring table on the right, red/yellow/green status chips, short notes.
Style/medium: realistic SaaS dashboard screenshot mockup, compact and work-focused.
Composition/framing: full browser window, dense but readable controls.
Lighting/mood: neutral productivity app, no decorative blobs.
Color palette: white, charcoal, slate, green, amber, red.
Text (verbatim): "VISUAL QA RUBRIC", "Text accuracy", "Object count", "Layout match", "Lighting", "Safety", "PASS", "FIX", "RETRY".
Constraints: no real company logos, no watermark, keep all UI text legible, avoid oversized hero style.

Use case: photorealistic-natural
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a 16:9 "SAME ROOM, FOUR HOURS" concept image showing one small reading corner rendered consistently at morning, noon, dusk, and night in four vertical slices. This demonstrates scene consistency, lighting/time variation, and spatial memory without changing furniture.
Scene/backdrop: cozy apartment reading nook with one armchair, side table, plant, bookshelf, window, and mug.
Subject: identical room layout across four time states.
Style/medium: photorealistic interior photography, natural textures.
Composition/framing: one seamless room divided into four vertical time bands, same camera angle.
Lighting/mood: morning soft sun, noon bright daylight, dusk warm orange, night lamp glow.
Color palette: varied by time while furniture stays constant.
Text (verbatim): small unobtrusive labels only: "MORNING", "NOON", "DUSK", "NIGHT".
Constraints: keep furniture positions identical, no people, no logos, no watermark, no fantasy elements.

Use case: product-mockup
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a 16:9 "REPAIR MANUAL MAKER" concept image: a fictional exploded repair guide for a compact desk lamp, with numbered parts, arrows, callouts, tiny tool illustrations, and a final assembled lamp thumbnail. This demonstrates structured visual reasoning and useful diagrams from plain language.
Scene/backdrop: clean workbench with the manual page floating above the real lamp parts.
Subject: desk lamp components: shade, hinge, spring, base, cable, switch, screws.
Style/medium: premium technical illustration blended with realistic product photography.
Composition/framing: centered exploded view with crisp callout lines and small labels.
Lighting/mood: bright studio lighting, practical, maker-lab feeling.
Color palette: off-white paper, graphite lines, safety orange highlights, brushed steel parts.
Text (verbatim): "REPAIR MANUAL MAKER", "1 SHADE", "2 HINGE", "3 SPRING", "4 BASE", "5 SWITCH".
Constraints: fictional product only, no brand logos, no watermark, no unsafe electrical instructions, prioritize legible labels.

Use case: infographic-diagram
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a 16:9 polished visual concept titled "SIGNAGE MAZE" showing a fictional museum wayfinding test: crisp multilingual signs in English, Chinese, Japanese, and Spanish guide visitors through four exhibit zones. The image demonstrates accurate text rendering, spatial layout, and visual hierarchy.
Scene/backdrop: bright modern museum atrium with overhead map, wall arrows, floor decals, and kiosk screen.
Subject: readable signage system with zone names: "入口 Entrance", "光影 Light", "城市 City", "出口 Exit".
Style/medium: realistic editorial design render, clean architectural visualization.
Composition/framing: wide view with signs distributed through the space, no giant poster board, signs integrated into the environment.
Lighting/mood: daylight, calm, precise, professional.
Color palette: white architecture, teal arrows, amber zone accents, dark graphite text.
Text (verbatim): "入口 Entrance", "光影 Light", "城市 City", "出口 Exit".
Constraints: no real brands, no logos, no watermarks, avoid unreadable gibberish; make text as legible as possible.

Use case: stylized-concept
Asset type: GPT Image 2 gallery derived play modes board, 2K landscape contact sheet
Primary request: Create a second polished 3x2 contact sheet of six NEW game-like play mechanics derived from the use cases in the companion board. These must be playable frameworks for experimenting with GPT Image 2 outputs, not repeats of earlier recorded mechanics such as product-to-flavor-lab, app-to-board-game, poster-to-mini-comic, Claim Court, Access Route Remix, Director's Cut Ladder, Aging Oracle, Taste Telemetry Duel, Edge-Case Hunt, prompt relay, constraint bingo, regression detective, variant draft arena, scene grammar kit, scale-shift microscope, wardrobe audit, gallery speedrun, drill-path puzzle, material truth table, waveform jam, or maintenance triage.
Panels:
1. Manual Maze: players receive a messy procedure card, generate a pocket manual, then race to spot missing warnings, unclear order, and ambiguous icons.
2. Price Canon Kitchen: players expand one receipt into a menu world; points reward price consistency, cuisine coherence, and believable supporting props.
3. Prop Continuity Poker: each player generates a different view of the same fictional object, then bids on which details survive across front, side, damaged, and hero shots.
4. Signage Babel Sprint: teams design multilingual wayfinding; judges test whether a traveler can infer route, exit, transfer, restroom, and help without reading all languages.
5. Figure Ladder Duel: players transform one dense chart into poster, slide, and thumbnail; each round removes 20 percent of visual clutter while preserving the claim.
6. Locked-Room Remix: players redesign an interior from a fixed reference layout; scoring favors style diversity while penalizing changed windows, wall geometry, or camera angle.
Style/medium: premium tabletop game documentation spread, realistic cards, score sheets, tokens, timers, judge stickers, mini output images, pencils, no real brands.
Composition/framing: 3 columns by 2 rows, six clearly separated panels; each panel shows the game components, scoring traces, and a small output sample.
Lighting/mood: warm studio tabletop lighting, collaborative design jam energy, precise enough for documentation.
Color palette: neutral table, varied accent colors for each game, crisp black rule text, colorful scoring tokens.
Constraints: no logos, no watermarks, no copyrighted characters; make each play mode visibly recordable with title, components, round structure, and scoring evidence; avoid repeating any prior gallery play format.

Use case: stylized-concept
Asset type: GPT Image 2 gallery research board, 2K landscape contact sheet
Primary request: Create a polished 3x2 visual research board showing six NEW practical and playful GPT Image 2 use cases, focused on capabilities reported in current GPT Image 2/API materials: high-quality generation and editing, flexible image sizes, high-fidelity image inputs, stronger instruction following, structured layouts, and reliable in-image text. Avoid repeating earlier gallery themes such as Cloud Oolong packaging, Route Bento app, midnight bookstore poster, Drill learning UI, tea brewing diagram, local legend store style transfer, weather portraits, recipe storyboards, museum field guides, prompt relay, constraint bingo, or regression detective.
Panels:
1. Micro-Manual Builder: turn a messy handwritten process into a clean illustrated pocket manual with numbered steps, safety marks, and small diagrams.
2. Receipt-to-Menu World: transform a plain restaurant receipt into a full fictional menu board, food photography, staff badge, and table tent while keeping prices coherent.
3. Continuity Prop Bible: generate a film prop reference sheet showing the same imaginary object from front, side, back, damaged, and hero-shot views.
4. Multilingual Wayfinding Kit: create coordinated transit signs in Chinese, English, Japanese, Korean, and icon language, testing legibility and layout hierarchy.
5. Research Figure Remixer: convert a dense academic chart into three reader-friendly visual explanations: poster, slide, and annotated thumbnail.
6. Input-Fidelity Interior Swap: use a room photo as a locked layout reference, then show three furniture/material redesign options without changing windows, walls, or camera angle.
Style/medium: premium design-research wall, realistic printed boards, image tiles, UI fragments, diagrams, mockups, small annotation labels, no real brands.
Composition/framing: 3 columns by 2 rows, each panel has a distinct mini-scene and a short title; title text should be simple and legible but the image must not depend on perfect typography.
Lighting/mood: bright studio table and pinboard lighting, crisp shadows, exploratory but production-ready.
Color palette: varied accents per panel, neutral paper, black ink, a few saturated markers.
Constraints: no logos, no watermarks, no real company marks, no copyrighted characters; make every panel feel like a recordable use case with visible before/after or source/output relationship.

Use case: stylized-concept
Asset type: gallery contact sheet for derived GPT Image play formats
Primary request: Create a polished 3x2 contact sheet showing six NEW game-like play formats derived from the use cases in the companion board. These are interaction/game mechanics for using GPT Image outputs, not simple repeats of earlier recorded games such as regression detective, constraint bingo, prompt relay, variant draft arena, scene grammar kit, scale-shift microscope, wardrobe audit, gallery speedrun, drill-path puzzle, material truth table, waveform jam, or maintenance triage.
Panels:
1. Claim Court: players prosecute or defend generated ad posters by placing evidence tokens on visual claims, missing caveats, and emotional manipulation cues.
2. Access Route Remix: players redraw a generated street scene for different bodies and contexts, scoring friction removed per edit round.
3. Director's Cut Ladder: players convert a script scene into increasingly constrained shot boards, then vote on the clearest emotional beat.
4. Aging Oracle: players predict which generated product details will fail first, then generate later-life variants to reveal outcomes.
5. Taste Telemetry Duel: two teams turn the same flavor card into packaging, lighting, texture, and vessel images, then blind-rank sensory fit.
6. Edge-Case Hunt: players draft rare generated dataset examples and build the smallest set that covers the most failure modes.
Style/medium: premium tabletop rulebook spread meets design-research wall, realistic game components, tokens, cards, mini boards, pencils, sticky notes, image tiles, no brand names.
Composition/framing: 3 columns by 2 rows, six clear panels; each panel contains a compact play setup with components and scoring traces; small labels okay but do not rely on exact typography.
Lighting/mood: warm studio table lighting with crisp shadows, collaborative and experimental.
Color palette: varied panel accents, neutral tabletop, black ink, colored tokens, photographic tiles.
Constraints: no logos, no watermarks, no real company marks, avoid repeating prior recorded play formats; every panel should feel playable and recordable.

Use case: stylized-concept
Asset type: gallery contact sheet for GPT Image use case research
Primary request: Create a polished 3x2 contact sheet showing six NEW practical GPT Image-family use cases that are playful but not repeats of product remix, museum field guides, UI portals, weather portraits, recipe still lifes, micro-brands, masked edits, character episodes, diagram dioramas, annotation archaeology, localization stress tests, RPG rarity ladders, data anomaly cards, assembly manuals, patent sketches, itinerary maps, facade lighting, board games, wardrobe boards, exhibit rehearsals, emergency drills, material substitution, waveform stages, or maintenance logs.
Panels:
1. Policy red-team ad lab: a wall of synthetic posters being checked for claims, disclosures, and misleading visual cues.
2. Accessibility empathy simulator: one street corner visualized through low vision, glare, one-handed use, wheelchair route, and audio-only cue overlays.
3. Script-to-camera blocking: a short dialogue scene transformed into overhead blocking, lens cards, actor marks, and lighting setup.
4. Future wear forecast: one everyday backpack shown as pristine, six-month scuffed, repaired, patched, and end-of-life material states.
5. Flavor-to-form atlas: a drink flavor profile turned into color, texture, vessel shape, garnish, and mood boards.
6. Synthetic dataset gap filler: a QA table showing rare edge-case product photos generated under unusual angles, weather, occlusion, and lighting.
Style/medium: high-end editorial design board, crisp photorealistic miniature scenes mixed with clean annotation graphics, no brand names.
Composition/framing: 3 columns by 2 rows, each panel visually distinct, small legible title areas but avoid relying on exact text; rich details, balanced margins.
Lighting/mood: bright studio-documentary lighting, practical and inventive.
Color palette: varied but coherent; white workspace, signal colors, natural material accents.
Constraints: no logos, no watermarks, no real company marks, avoid repeating earlier gallery concepts; prioritize clear use-case differentiation.

Create a second single high-quality landscape concept board showing six NEW derived game-like GPT Image 2 play formats inspired by the previous use cases. Avoid repeating all previous play formats: hidden-object hunt, brand slot machine, moodboard battle, time-machine photo chain, character passport, prompt recipe kitchen, toy remixes, micro-brand kits, old photo repair, regression detective, constraint bingo, prompt relay telephone, variant draft arena, scene grammar kit, scale-shift microscope, plus the six use cases from the previous board. 

No readable text except tiny abstract marks. Style: polished tabletop/game design artifact plus creative operations studio, six-panel grid, energetic but organized, clear interaction mechanics, balanced multi-color palette, no logos, no famous people, no copyrighted characters.

Six fresh derived play formats:
1. Continuity wardrobe audit: players receive generated outfit boards and score whether fabric, wear, season logic, and packing constraints stay consistent across image rounds.
2. Gallery layout speedrun: players race to arrange generated mini-exhibits under constraints like visitor path, lighting budget, object fragility, and wall density.
3. Drill-path puzzle: generated floor plans and scene cards become a cooperative route-planning puzzle where players place blockers, exits, and assembly markers.
4. Material truth table: one object is generated in many materials; players rank which version best communicates weight, softness, durability, cost, and repairability.
5. Waveform stage jam: players convert abstract sound cards into stage scene prompts, then compare the resulting lighting, props, and movement language.
6. Maintenance triage draft: generated machine-inspection cards are drafted into urgent, watch-list, and safe piles based on visible wear, residue, missing parts, and tool evidence.

Make it look like a reusable play kit for evaluating and playing with image generation, with each panel self-explanatory visually.

Create a single high-quality landscape concept board showing six NEW playful GPT Image 2 use cases. Avoid repeating these previous themes: toybox product remix, alternate-history field guide, UI-to-world portal, emotion weather portraits, recipe still life, micro-brand launch kit, masked reality patch, character consistency episode, diagram-to-diorama, annotation archaeology, localization stress test, game asset rarity ladder, data outlier evidence cards, assembly manual, patent sketch simulator, itinerary memory map, building lighting rehearsal, board-game prototype factory, regression detective, constraint bingo, prompt relay, variant draft, scene grammar kit, scale-shift microscope. 

No readable text except tiny abstract marks. Style: polished creative lab research board, six-panel grid, crisp realistic objects with subtle editorial illustration polish, vivid balanced palette, no logos, no famous people, no copyrighted characters.

Six fresh panels:
1. Wardrobe continuity lab: a capsule wardrobe is turned into a season-by-season outfit continuity board for a fictional character, with fabric swatches, silhouettes, laundry wear, and travel packing variants, but without relying on face identity.
2. Museum exhibit rehearsal: a curator prompt becomes a miniature gallery installation with plinths, wall lighting, visitor flow, object labels represented only as abstract marks, and alternate display layouts.
3. Emergency drill simulator: a floor plan becomes realistic staged safety-drill scenes with signage shapes, blocked paths, assembly points, and before/after readiness photos.
4. Material substitution studio: one product form is explored in glass, paper pulp, recycled plastic, ceramic, woven fabric, and metal, preserving the silhouette while changing tactile qualities.
5. Sound-to-scene composer: waveform-like audio patterns become a physical stage design with lights, props, motion trails, and mood zones for an imaginary performance.
6. Maintenance log visualizer: routine machine notes become a visual checklist wall with worn parts, oil traces, inspection stamps, tool shadows, and replacement candidates.

Make each panel self-explanatory as a practical GPT Image 2 experiment, not just a pretty picture.

Create a single high-quality 2K landscape concept board showing six NEW derivative game-like play formats for GPT Image 2. Avoid repeating hidden-object treasure hunt, brand slot machine, moodboard battle, time-machine photo chain, character passport, prompt recipe kitchen, toy remixes, micro-brand kits, and old photo repair. No readable text except abstract symbols. Style: polished game/product design artifact, clean six-panel grid, energetic interaction, vivid balanced palette.

Six panels:
1. Regression detective: two near-identical generated scenes side by side with players marking changed details, lighting drift, missing props, and style inconsistencies.
2. Constraint bingo studio: a bingo-like board of visual constraints such as camera angle, material, era, weather, palette, and composition, with image tiles filling completed rows.
3. Prompt relay telephone: a chain of players pass only an image forward; each station reinterprets it, creating a visible evolution trail from object to poster to scene to interface.
4. Variant draft arena: multiple generated options are laid out like collectible draft picks, and players choose parents to breed the next round of visual variants.
5. Scene grammar kit: modular cards for subject, action, environment, lens, lighting, texture, and mood snap together physically to produce one coherent final scene.
6. Scale-shift microscope: one generated object is explored across macro, normal, miniature, and world-size views, with magnifying instruments and miniature figures for comparison.

Make it look like a reusable playbook for creative experiments, with each panel self-explanatory, no brand logos, no famous people, no copyrighted characters, no exact readable typography.

Create a single high-quality 2K landscape concept board showing six NEW playful GPT Image 2 use cases that avoid character consistency, micro-brand launches, toybox product remixes, old photo repair, recipe still lifes, surreal masked patches, UI-to-world portals, and emotion weather portraits. No readable text except tiny abstract marks. Style: polished research lab concept art, clean six-panel grid, vivid but balanced palette, crisp realistic objects with subtle illustration polish.

Six fresh panels:
1. Data outlier evidence cards: a spreadsheet/CSV dashboard becomes a wall of visual forensic cards, each card showing a concrete scene explaining an anomaly.
2. Parts-to-assembly manual: scattered hardware parts on a workbench become an elegant exploded-view instruction poster and final assembled gadget.
3. Patent sketch simulator: a rough invention description becomes annotated-looking blueprint drawings, prototype photos, and test-rig scenes, with only abstract pseudo-labels.
4. Itinerary memory map: a travel schedule becomes a tactile tabletop map with miniature landmarks, route threads, tickets, weather icons, and day-by-day tokens.
5. Lighting rehearsal: one building facade is previewed under several event lighting states, rain/dusk/festival/emergency, all consistent geometry.
6. Board-game prototype factory: a theme paragraph becomes a playable tabletop board, tokens, encounter cards, dice, and miniature terrain.

Make every panel self-explanatory, no brand logos, no famous people, no copyrighted characters, no exact readable typography.

Create a second single high-quality 2K landscape concept board showing six derivative game-like GPT Image 2 play formats inspired by the first set. No readable text except simple abstract symbols. Style: playful but polished product concept art, clean panels, energetic interaction, vivid non-monochrome palette, no logos, no famous people, no copyrighted characters.

Six panels:
1. Visual treasure hunt: a generated city scene contains hidden whimsical objects, with magnifying-glass UI overlays and players pointing at clues.
2. Brand remix slot machine: a fictional snack package cycles through packaging styles, photo scenes, and poster formats like a creative slot machine.
3. Moodboard battle cards: two players compare generated fashion/interior mood cards on a tabletop, each card has coherent color/material props.
4. Time-machine photo chain: one family snapshot becomes three versions across eras and genres, shown as a branching timeline wall.
5. Character passport: an original mascot character appears in passport-stamp pages across different worlds while staying visually consistent.
6. Prompt recipe kitchen: ingredients like color chips, camera lenses, layout tiles, and emotion tokens are mixed in a chef-like studio to generate a final poster.

Make it look like a record of product-play experiments: clear visual ideas, each panel self-explanatory, cohesive grid, no exact readable typography.

Create a single high-quality 2K landscape concept board showing six playful GPT Image 2 use cases as separate clean panels in one cohesive image. No readable text except tiny abstract UI marks. Style: polished product-design concept art, crisp lighting, colorful but balanced, modern studio quality, realistic objects with subtle illustration polish.

Panel ideas:
1. A messy hand-drawn napkin sketch transforming into a polished visual explainer poster on a desk.
2. A consistent original cartoon explorer character appearing across a four-frame mini comic strip, same outfit and face.
3. A small indie beverage product shown in three ad variants: studio shot, outdoor lifestyle shot, and poster layout.
4. An apartment room evolving into three interior design options with consistent furniture.
5. A wireframe app screen becoming a refined mobile UI mockup.
6. An old travel photo being cleaned up, colorized, and extended into a cinematic memory scene.

Make the six panels feel like a useful research artifact, with clear visual separation, no brand logos, no famous people, no copyrighted characters, no artist-name style imitation.

Create a second polished 2x2 contact sheet of derived GPT image generation play patterns, no watermark, minimal or no readable text, four square panels divided by thin white gutters:
Panel A: Recipe-to-prop cinema still - an overhead cinematic detective-table still life where a recipe card becomes clues: flour as desert dunes, tomatoes as red planets, basil as forest, measuring spoons as instruments, warm film lighting.
Panel B: Micro-brand launch kit - a coherent fictional pop-up beverage brand shown as one compact launch board: hero can, packaging, small storefront vignette, ad poster, social tile mockup; energetic but not cluttered.
Panel C: Masked reality patch - a mundane kitchen counter scene where one rectangular edited region is replaced by a perfectly lighting-matched surreal portal garden, with subtle mask-edge cues like before/after design review.
Panel D: Character consistency mini episode - four tiny comic frames inside one panel featuring the same small explorer character with the same red scarf and brass lantern across morning, noon, dusk, night locations; consistent outfit and silhouette.
Make each panel visually different, production-quality, playful experimentation mood, crisp 2x2 sheet.

Create a polished 2x2 contact sheet titled visually by panel composition only, no large captions, no watermark. Four square panels separated by thin white gutters, each showing a distinct fun GPT image generation use case:
Panel A: Toybox product remix - a realistic studio shelf photo of ordinary desk objects reimagined as collectible blind-box toys, with charming packaging, tiny mascot tags, and colorful rarity-card energy.
Panel B: Alternate-history field guide - a museum archive plate showing impossible future fossils and strange tools, precise natural-history illustration mixed with believable catalog photography.
Panel C: UI-to-world portal - a software dashboard transformed into a physical control room installation, buttons as chunky machinery, charts as glowing architectural panels, cinematic but readable.
Panel D: Emotion weather portrait - a stylized portrait where a calm face contains tiny weather systems, sunbreaks, rain curtains, and pressure-map lines, editorial poster quality.
Use varied palettes across panels, high detail, playful but sophisticated, square contact sheet, crisp composition.

Create a second polished 3x2 visual research board titled "DERIVED PLAY MODES". Six experimental GPT-image-2 gameplay/use-case concepts, each with a short readable English label and a small Chinese subtitle. Panel 1: "Prompt Escape Room" showing a puzzle poster where hidden visual clues unlock the next prompt. Panel 2: "Brand Multiverse" showing the same fictional tea brand across cyberpunk, museum, and grocery shelf identities. Panel 3: "Before/After Time Machine" showing one street corner across 1926, 2026, and 2126. Panel 4: "UI Fiction Simulator" showing imaginary app screens for an impossible product, realistic but fictional. Panel 5: "Recipe to Comic" showing a cooking recipe transformed into a 4-panel comic with readable step labels. Panel 6: "Worldbuilding Atlas" showing a fantasy transit map, stamps, product labels, and field notes for one fictional city. Contemporary design research board, crisp layout, strong but not chaotic color, readable labels, original fictional content only, no real logos, no celebrity likenesses.

Create a polished 3x2 visual research board titled "GPT-IMAGE-2 PLAYBOOK". Six distinct panels, each panel has a short readable English label and a small Chinese subtitle. Panel 1: multilingual poster typography with crisp Chinese/Japanese/Hindi text blocks. Panel 2: product packaging mockup for a fictional sparkling tea can, no real brands. Panel 3: UI mockup of a fictional analytics app on a tablet. Panel 4: multi-panel manga/storyboard with consistent original character. Panel 5: educational infographic explaining urban rainwater recycling with clean arrows and labels. Panel 6: interior design plan showing the same tiny apartment in three styles. Contemporary editorial design, vibrant but balanced palette, sharp typography, clean grid, high detail, no real brand logos, no celebrity likenesses.

Use case: stylized-concept
Asset type: derived-play mechanics board, 2x2 visual grid for GPT Image 2 experiments
Primary request: Create a second polished visual board that derives four repeatable play mechanics from the first use-case research. No brand logos. Minimal readable labels only.
Panel 1: "Remix Dice" - six dice faces trigger visual transformations of the same object: material, era, mood, camera angle, audience, and constraint; show one backpack transformed by the dice results.
Panel 2: "Continuity Quest" - a character bible sheet for one recurring mascot across pose turnaround, expressions, costume swaps, and prop variants; emphasize consistency.
Panel 3: "Prompt Arcade" - a playful control panel where sliders and buttons generate image variants for product colorway, lighting, texture, and layout; show thumbnails emerging from the machine.
Panel 4: "Edit Relay" - a step-by-step relay race of image edits: sketch -> rendered prop -> packaging -> ad poster -> game card, each step preserving key visual DNA.
Composition: 2x2 grid with thin neutral dividers, coherent across panels but each mechanic distinct.
Style: energetic, maker-lab aesthetic, tactile objects, clean lighting, high-resolution square image, no clutter, no watermark.

Use case: productivity-visual
Asset type: exploration board, 2x2 visual grid for GPT Image 2 use cases
Primary request: Create a polished visual research board showing four fun GPT Image 2 use cases as distinct panels. No brand logos. Minimal readable labels only.
Panel 1: "Product Lab" - a fictional soda can packaging design evolving into three shelf-ready mockups, crisp studio lighting.
Panel 2: "Story Remix" - the same tiny astronaut character appearing in three comic beats across different art styles, consistent silhouette.
Panel 3: "Learning Lens" - a colorful exploded-view educational diagram of a mechanical pencil, clean arrows and simple callouts.
Panel 4: "Game Kit" - a tabletop game asset sheet with cards, tokens, terrain tiles, and item icons, coherent fantasy-adventure style.
Composition: 2x2 grid with thin neutral dividers, each panel visually rich and self-contained, high-end editorial design, no clutter.
Style: sharp, modern, playful but practical, strong material detail, balanced color palette, square image.

Use case: productivity-visual
Asset type: comparison matrix poster
Primary request: Create a visual experiment sheet for a game called "Constraint Bingo". It compares the same concept under eight prompt constraints in a neat grid.
Scene/backdrop: design-lab pinboard with printed mini images.
Subject: eight small images of the same object, a red bicycle, transformed by constraints: macro, isometric, noir, watercolor, pixel, product photo, blueprint, cinematic.
Text requirements: title exactly "Constraint Bingo"; eight labels exactly "macro", "isometric", "noir", "watercolor", "pixel", "product", "blueprint", "cinematic".
Style: clean comparison board, readable small labels, visually distinct variants, professional creative workshop artifact.
Composition: 4x2 grid, title at top, no extra text, no watermark.

Use case: stylized-concept
Asset type: sprite sheet / asset pack
Primary request: Create a compact 3x3 asset sheet for a cozy image-generation mini game. Each tile is a different reusable visual token: camera, scissors, magic brush, dice, seed, portal, speech bubble, palette, checkmark badge.
Scene/backdrop: flat clean asset sheet, transparent-looking checkerboard printed behind tiles but final image can be opaque.
Subject: nine consistent game icons with the same lighting, outline weight, and material style.
Text requirements: no text.
Style: polished 3D clay + enamel icons, friendly, tactile, consistent design system, game-ready.
Composition: 3x3 grid, equal padding, each icon centered, no watermark, no labels.

Use case: illustration-story
Asset type: four-panel sequential story image
Primary request: Create a single image containing four comic panels for a "same prompt evolves" game. The base idea is a tiny robot gardener finding a strange seed, then each panel changes one constraint.
Scene/backdrop: compact garden workshop.
Subject: same tiny robot gardener kept visually consistent across all four panels.
Text requirements: small panel labels exactly "BASE", "MOOD", "ERA", "GENRE". No speech bubbles.
Style: charming cinematic comic, strong continuity, clear panel borders, expressive lighting shifts.
Composition: 2x2 grid, each panel visually distinct but same character design, no watermark, no extra text.

Use case: illustration-story
Asset type: game card prototype
Primary request: Create one illustrated tabletop game card for a prompt-generation game. The card should invite players to mutate an image concept.
Scene/backdrop: single card on a neutral table, photographed from above.
Subject: a polished fantasy-tech card with icon art showing a glowing dice rolling across tiny image thumbnails.
Text requirements: card title exactly "REMIX DICE"; rule text exactly "Change one noun, one style, one constraint."; footer exactly "Play 02".
Style: premium board-game component, readable typography, tactile paper texture, fun but refined.
Composition: portrait card centered, full card visible, no other cards, no watermark, no extra text.

Use case: scientific-educational
Asset type: classroom explainer infographic
Primary request: Create an accurate, playful explainer poster about how an image prompt becomes a visual output, suitable for a creative coding workshop.
Scene/backdrop: clean whiteboard-style infographic with small illustrations.
Subject: a four-step flow from prompt to planning to rendering to review.
Text requirements: title exactly "Prompt → Plan → Pixels"; four step labels exactly "1 Prompt", "2 Constraints", "3 Generate", "4 Review"; include no other long text.
Style: clear educational infographic, colorful icons, crisp arrows, readable English labels, high legibility.
Composition: portrait 4:5 poster, structured diagram, ample spacing, no fake tiny paragraphs, no watermark.

Use case: ui-mockup
Asset type: app interface mockup
Primary request: Create a high-fidelity dashboard mockup for an image experimentation app named "Playground Ledger" that records prompts, generated variants, scores, and reuse warnings.
Scene/backdrop: direct screenshot-style UI, no laptop frame.
Subject: dense but elegant SaaS dashboard with left navigation, prompt cards, variant grid, and a right-side "avoid repeats" panel.
Text requirements: app title exactly "Playground Ledger"; panel headings exactly "Use Cases", "Variants", "Avoid Repeats", "Next Plays". Use English UI copy only and keep it short.
Style: production-grade SaaS UI, neutral base with strategic accent colors, 8px card radius, readable labels, realistic data table.
Composition: 16:10 desktop screenshot, crisp typography, no marketing hero, no extra explanatory text, no watermark.

Use case: product-mockup
Asset type: product concept render
Primary request: Design a fictional "Prompt Potion" soda can family where each flavor represents an image-generation mode: Render, Edit, Remix, Story.
Scene/backdrop: clean retail shelf with four cans aligned, small swatch cards and ingredient-style prompt fragments around them.
Subject: four premium illustrated beverage cans, no real logos.
Text requirements: can labels should read exactly "RENDER", "EDIT", "REMIX", "STORY"; shelf label exactly "Prompt Potion".
Style: high-end playful product photography, sharp packaging, colorful but disciplined palette, realistic reflections.
Composition: landscape product shot, 4 cans visible front-on, strong lighting, readable labels, no extra brand text, no watermark.

Use case: ads-marketing
Asset type: square social poster concept
Primary request: Create a playful Chinese poster for a fictional AI image lab event called "图像实验夜". The image should demonstrate strong text rendering and structured design.
Scene/backdrop: bright studio tabletop with tiny printed photo strips, stickers, prompt cards, and a small monitor showing abstract image thumbnails.
Subject: a clean, editorial event poster composition, not a real brand.
Text requirements: large main title exactly "图像实验夜"; smaller subtitle exactly "8个玩法 · 2组实验"; bottom tag exactly "GPT Image Playbook".
Style: polished modern editorial poster, crisp Chinese typography, balanced hierarchy, energetic but not cluttered.
Composition: square 1:1, main title centered, props arranged around edges, enough negative space, no extra text, no watermark.

Use case: infographic-diagram
Asset type: derived play concept image
Primary request: Create a 16:9 concept image called "IMAGE QUEST MAP" showing a branching map for turning one generated image into a mini project. Nodes must be readable and labeled: IDEA, MOODBOARD, HERO, VARIANTS, ASSETS, MOCKUP, SHARE. Use a polished visual strategy workshop wall with sticky notes, printed thumbnails, arrows, and a central map. Sophisticated creative operations feel, no real brands, no watermark.

Use case: product-mockup
Asset type: derived play concept image
Primary request: Create a 16:9 image of a playful "ONE OBJECT, MANY LIVES" challenge. Show the same simple ceramic mug transformed into six coherent product identities arranged in a grid: cafe mug, camping mug, museum gift mug, kids art mug, premium corporate mug, sci-fi prop mug. Keep the mug silhouette visibly consistent while styling changes. Add small readable labels: CAFE, CAMP, MUSEUM, KIDS, OFFICE, FUTURE. Studio product photography, no real brands, no watermark.

Use case: illustration-story
Asset type: derived play concept image
Primary request: Create a 16:9 visual of a game called "VISUAL TELEPHONE" where one image mutates through five steps across a gallery wall. Start with a simple red bicycle, then evolve into a delivery bike, then a floating market stall, then a tiny city airship, then a whimsical poster. Include readable labels: STEP 1, STEP 2, STEP 3, STEP 4, STEP 5. Make it feel like a creative workshop activity, coherent lighting, no real brands, no watermark.

Use case: productivity-visual
Asset type: derived play concept image
Primary request: Create a 16:9 board-game style visual called "PROMPT REMIX CARDS". Show a tabletop with colorful cards that players combine to make image prompts. Cards must have readable category titles: SUBJECT, STYLE, CONSTRAINT, TWIST, FORMAT. Include sample cards: "lamp", "clay render", "no text", "tiny city", "poster". Polished product photography, playful but clean, no real brands, no watermark.

Use case: ui-mockup
Asset type: concept image for a GPT Image use-case research board
Primary request: Create a sharp 16:9 high-fidelity UI mockup image of a fictional visual editing app called "Canvas Lab". The screen shows a left tool rail, central image canvas, right inspector, and timeline at bottom. It demonstrates multi-turn image editing with readable UI labels: Generate, Edit, Mask, Compare, Export. Quiet professional SaaS style, realistic desktop app screenshot, no real company logos, no watermark.

Use case: scientific-educational
Asset type: concept image for a GPT Image use-case research board
Primary request: Create a clean 16:9 educational explainer poster titled "HOW A HEAT PUMP MOVES HEAT". Show a simplified house, outdoor unit, indoor coil, arrows for heat flow, and four labeled steps with accurate short text: 1 EVAPORATE, 2 COMPRESS, 3 CONDENSE, 4 EXPAND. Use crisp infographic styling, accurate directional arrows, readable typography, calm professional palette, no real brand logos, no watermark.

Use case: illustration-story
Asset type: concept image for a GPT Image use-case research board
Primary request: Create a 16:9 illustrated character continuity sheet for a fictional tiny astronaut named NOVA across six panels: front pose, side pose, surprised face, running pose, sitting pose, and bedtime scene. The same helmet shape, red scarf, star patch, and proportions must stay consistent in every panel. Clean animation-production style, labels under panels with accurate text: FRONT, SIDE, FACE, RUN, SIT, STORY. No watermark.

Use case: ads-marketing
Asset type: derived GPT Image 2 play concept, single finished image
Primary request: Create a fictional local pop-up event poster derived from “turn real-world objects into playful campaign material”.
Scene/backdrop: A lively but realistic community noticeboard with one central poster pinned up, surrounded by small paper tickets and sticky notes.
Subject: Central poster advertises an imaginary weekend event: “TINY MAKER MARKET”. Visual concept: everyday objects turned into miniature booths, like a matchbox as a stall, a bottle cap as a stage, and a spoon as a bridge.
Text: Include exact readable English copy: “TINY MAKER MARKET”, “APR 26”, “10 AM - 4 PM”, “Make small things feel big”.
Style: whimsical photorealistic miniature diorama poster, tactile paper texture, clear hierarchy, inviting but not childish.
Composition: vertical 4:5, poster centered, surrounding notes add context, no real venue or brand logos, no watermark.

Use case: ads-marketing
Asset type: concept image for a GPT Image use-case research board
Primary request: Create a polished 16:9 visual concept showing a fictional instant product-campaign generator: a desk scene with three coherent beverage can ad variants on a wall, each with accurate short readable text: "CRISP", "ZERO", "SPARK". Show the same can identity adapted into social post, shelf talker, and outdoor poster mockups. Modern studio lighting, realistic materials, high-end creative director mood, no real brands, no watermark.

Use case: scientific-educational
Asset type: derived GPT Image 2 play asset sheet, single finished image
Primary request: Create a printable tabletop learning game prototype about food chains, showing how GPT Image 2 can make educational play materials.
Scene/backdrop: A neatly arranged print-and-play game kit on a classroom table.
Subject: Include a small foldable board, 12 illustrated cards, tokens, and a rule card. Theme: pond food chain. Use simple readable labels on cards: “sun”, “algae”, “snail”, “fish”, “frog”, “heron”, “energy”, “storm”.
Text: Rule card title should read “POND CHAIN”. Keep other text minimal and readable.
Style: friendly educational illustration mixed with realistic printed paper mockup, accurate simple ecology icons, not childish clutter.
Composition: top-down 4:5, all pieces visible, no watermark, no real school logos.

Use case: product-mockup
Asset type: derived GPT Image 2 play asset sheet, single finished image
Primary request: Create a fictional product packaging variant system derived from the YUZU POP ad, showing fast flavor exploration.
Scene/backdrop: Clean studio shelf with six cans arranged in a row plus a small flavor legend card.
Subject: Six fictional sparkling tea cans with consistent layout and distinct flavors: YUZU POP, PEACH FIZZ, LIME SNAP, BERRY LIFT, PLUM GLOW, GINGER ZING. Each can should have simple fruit graphics and color-coded accents while staying in one brand system.
Text: Flavor names should be readable on the cans and legend card. No real brand marks.
Style: premium CPG packaging design board, realistic metal cans, subtle reflections, bright but controlled palette.
Composition: horizontal 16:9 design review board, every can fully visible, no watermark.

Use case: stylized-concept
Asset type: derived GPT Image 2 play asset sheet, single finished image
Primary request: Create a reusable character asset sheet derived from the mini comic robot, showing a consistent original robot mascot for sticker/game/social use.
Scene/backdrop: Clean off-white studio background with subtle grid marks.
Subject: A small round delivery robot mascot with a blue scarf and expressive screen eyes. Show 8 poses/emotions in a neat 4x2 grid: happy, thinking, scanning, carrying a book, waving, surprised, sleepy, celebrating. Keep the robot design consistent across all poses.
Text: Small readable labels only: “happy”, “think”, “scan”, “carry”, “wave”, “wow”, “sleep”, “yay”.
Style: polished 3D-toy illustration, soft shadows, asset-sheet clarity, no real franchise influence, no watermark.
Composition: square asset sheet, generous spacing, every pose fully visible.

Use case: ui-mockup
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a fictional mobile app interface mockup for “Mood Kitchen”, showing how GPT Image 2 can quickly produce high-fidelity UI concepts.
Scene/backdrop: A realistic smartphone lying on a neutral kitchen counter with herbs and a spoon nearby.
Subject: The phone screen displays a clean recipe app UI with tabs, recipe cards, ingredient chips, and a calm “today's mood” selector. Use readable UI text: “Mood Kitchen”, “Calm”, “Quick”, “Spicy”, “15 min”, “Ginger noodle bowl”, “Save”.
Style: modern utilitarian consumer app, high fidelity UX mockup, not a marketing landing page, restrained colors with green, tomato red, cream, charcoal.
Composition: vertical 4:5, phone screen large and legible, real-world context, no real logos, no watermark.

Use case: illustration-story
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a 4-panel original mini comic that demonstrates consistent character and sequential storytelling.
Scene/backdrop: A cozy futuristic neighborhood library at night.
Subject: Original character: a small round delivery robot with a blue scarf and expressive screen eyes, helping a child find a missing moon-shaped bookmark. Four panels, clear panel borders, consistent robot design, cinematic but readable.
Text: Minimal readable English captions only: Panel 1 “Lost?”, Panel 2 “Scan”, Panel 3 “Found!”, Panel 4 “Story time”.
Style: warm original graphic novel illustration, not in the style of any specific living artist or known franchise, charming lighting, rich details.
Composition: square page, 2x2 panel layout, no watermark.

Use case: ads-marketing
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a fictional snack brand campaign image showing how GPT Image 2 can produce a product ad with clean typography.
Scene/backdrop: A colorful studio tabletop with a fictional sparkling tea can named “YUZU POP”, citrus peel spirals, ice cubes, tiny bubbles, and condensation.
Subject: One hero can and two background cans. Include exact readable English copy only: “YUZU POP”, “Citrus Sparkling Tea”, “Zero Sugar”, “Bright. Crisp. Playful.”
Style: premium product photography mixed with crisp graphic ad layout, fresh yellow, teal, white, and small coral accents, no real brands.
Composition: vertical poster 4:5, product large in center, copy integrated as ad typography, polished commercial finish, no watermark.

Use case: productivity-visual
Asset type: exploratory GPT Image 2 use-case card, single finished image
Primary request: Create a polished Chinese visual use-case demo for “AI 一页纸旅行任务板”.
Scene/backdrop: A clean desk photographed from above, with a printed A4 travel planning board for a fictional weekend in Hangzhou, surrounded by a coffee cup, transit card, marker, and phone.
Subject: The paper board should show structured sections with readable Chinese headings: “天气”, “路线”, “预算”, “拍照点”, “备选计划”. Use simple neat Chinese labels, not long paragraphs. Make it feel like GPT Image 2 can generate text-rich practical planning visuals.
Style: photorealistic editorial product photography, natural morning light, crisp typography, practical and delightful.
Composition: top-down 4:5, the board centered and fully visible, no brand logos, no watermark.

Create a polished 2K visual board of six DERIVED play modes based on the four new GPT-image-2 use cases, single clean 3x2 grid poster, no logos, no copyrighted brands. Title text in simplified Chinese: "GPT-image-2 派生玩法 D". Six labeled panels with accurate Chinese labels: 1 "编辑接龙", 2 "回滚竞速", 3 "约束闯关", 4 "差异侦探", 5 "来源护照", 6 "资产谱系树". Visualize each panel: edit-chain relay where each player changes one visual variable; speed challenge to roll back a bad revision to the best prior state; constraint challenge cards requiring exact aspect ratio, text, palette, object count; detective board spotting visual differences across versions; a passport-like document recording prompt/reference/license/checks for an asset; a family tree of asset variants and descendants with version nodes. Modern editorial layout, white background, thin dividers, distinct accent colors, highly legible Chinese labels, crisp typography, professional product-design style, no watermarks.

Create a polished 2K visual research board showing four NEW non-repeated GPT-image-2 use cases, single clean 2x2 grid poster, no logos, no copyrighted brands. Title text in simplified Chinese: "GPT-image-2 新玩法 C". Four labeled panels with accurate Chinese labels: 1 "源图编辑链", 2 "多轮修订栈", 3 "视觉 QA 仪表盘", 4 "素材溯源板". Visualize each panel: a photo transformed through sequential edit steps with locked subject identity; a layered revision history of one generated scene with annotations and diff marks; a design QA dashboard comparing expected vs actual visual constraints with pass/fail chips; a provenance board showing prompt, reference thumbnails, edit steps, rights notes, and final asset lineage. Modern editorial layout, white background, thin black dividers, small colored accent bars, highly legible Chinese text, crisp typography, professional design, no watermarks.

Create a second polished 2K visual research board of six derived playful product/game ideas based on GPT-image-2 capabilities, single clean grid poster, no logos, no copyrighted brands. Title text in simplified Chinese: "GPT-image-2 派生玩法 B". Six labeled panels with accurate Chinese text labels: 1 "广告盲盒", 2 "角色连续剧", 3 "菜单翻译器", 4 "空间换装", 5 "地图寻宝", 6 "错图挑战". Visualize each panel: random product ad creative cards, a consistent original character across three episode thumbnails, a food menu transformed into bilingual appetizing visuals, an interior room with swappable style skins, an illustrated city treasure map with clues, and a spot-the-error challenge image with subtle differences. Modern editorial layout, white background, thin black dividers, distinct accent colors, highly legible Chinese labels, crisp typography, professional design, no watermarks.

Create a polished 2K visual research board showing six fun GPT-image-2 use cases as a single clean grid poster, no logos, no copyrighted brands. Title text in simplified Chinese: "GPT-image-2 玩法实验 A". Six labeled panels with accurate Chinese text labels: 1 "多语言海报", 2 "假想产品棚拍", 3 "App 截图原型", 4 "四格漫画", 5 "数据地图", 6 "电影感剧照". Each panel should look like a distinct high-quality example: a bilingual poster with crisp typography, a futuristic small gadget product photo, a mobile app mockup screen, a consistent character four-panel comic, an infographic city map, and a cinematic still. Modern editorial layout, white background, thin black dividers, small colored accent bars, highly legible text, professional design, no watermarks.

Create a second polished 2x2 contact sheet of DERIVED play modes from the four new GPT Image 2 use cases. Do not repeat generic ads, UI dashboards, infocards, comics, product kits, localization posters, visual exams, tabletop games, children's books, creator shot recipes, thumbnail ladders, maps, receipts, calendars, or worldbuilding kits. Overall style: premium applied-research board, clean margins, useful details, no real brands, no watermarks.

Panel A label: "17. Failure Drill Cards". Derived from Service Rehearsal. Visual: a set of operational drill cards for a clinic front desk: angry patient, missing appointment, broken kiosk, language barrier; each card shows trigger, staff response, success signal.
Panel B label: "18. Incident Postmortem Diorama". Derived from Architecture X-Ray. Visual: a miniature physical diorama of a software outage timeline, with service components, bottleneck glow, alert stream, rollback switch, and readable time markers.
Panel C label: "19. Feed DNA Mutator". Derived from Persona Style Bible. Visual: nine social post thumbnails for the same fictional science educator, varying only framing, prop, density, and emotional tone; include a small scoring strip for recognition vs novelty.
Panel D label: "20. Renovation Constraint Solver". Derived from Before-After Decision Tree. Visual: a room planning board where constraints are movable tokens: budget, renter-safe, video calls, storage, lighting, noise; three floor-plan thumbnails update under different token combinations.

Use case: productivity-visual
Asset type: derived non-repeated play-mode contact sheet
Primary request: derive second-order玩法 from the new group: training drills, incident storytelling, visual identity mutation, and constraint-solving visual planning.
Composition: 2x2 grid, labels 17-20, production-grade and inspectable, clear relationship to the source use case.
Avoid: real company branding, celebrity likenesses, copyrighted characters, generic repeated prior concepts, illegible gibberish, decorative-only visuals.

Create a polished 2x2 contact sheet of NEW, non-repeated GPT Image 2 playground use cases. Do not depict generic ads, UI dashboards, infocards, comics, product kits, localization posters, visual quizzes, tabletop games, children's books, creator shot recipes, thumbnail ladders, maps, receipts, calendars, or worldbuilding kits. Overall style: premium applied-research board, clean margins, inspectable details, no real brands, no watermarks.

Panel A label: "13. Service Rehearsal". Visual: a split-scene simulation of a small clinic reception during a rush hour, showing staff positions, customer emotions, queue friction points, accessibility concerns, and service recovery moments as realistic annotated storyboard stills.
Panel B label: "14. Architecture X-Ray". Visual: a software architecture concept translated into a layered physical cutaway machine: API gateway, worker queue, cache, database, observability pipes, failure points as tangible components, readable labels only for these system parts.
Panel C label: "15. Persona Style Bible". Visual: a social account visual personality system for a fictional science educator: profile avatar concepts, recurring framing rules, color accents, prop vocabulary, caption tone cards, without platform logos.
Panel D label: "16. Before-After Decision Tree". Visual: a home office improvement board that branches from one messy room into three realistic redesign paths: budget fix, ergonomic fix, creator studio fix; include material samples and tradeoff notes.

Use case: productivity-visual
Asset type: non-repeated use-case contact sheet
Primary request: draw a new group of GPT Image 2 use cases that avoid prior recorded directions and focus on service simulation, code architecture visualization, creator visual systems, and decision-tree planning.
Composition: 2x2 grid, each panel distinct and useful, crisp labels, realistic production quality.
Avoid: previously recorded concepts, real company branding, celebrity likenesses, copyrighted characters, illegible gibberish, decorative-only visuals.

Create a second polished 2x2 contact sheet derived from the first group's ideas, exploring playful variations of GPT Image 2 workflows. Overall style: premium experimental design board, clean margins, consistent labels, no brand logos, no watermarks.

Panel A label: "5. Recipe card remix". Visual: a vertical recipe card generated from the noodle poster concept, with an illustrated bowl, ingredient icons, and readable exact title "Midnight Noodle Kit" plus short clean steps.
Panel B label: "6. Dashboard to room". Visual: transform the finance dashboard idea into a physical command-center desk setup: monitor shows neat widgets, sticky notes, calendar blocks, readable labels "Budget", "Focus", "Save".
Panel C label: "7. Character sheet". Visual: the same red delivery robot from the comic as a game-ready character sheet: front, side, back, expressions, prop teacup, color swatches, compact notes.
Panel D label: "8. Ad campaign triptych". Visual: three small product-ad variations for the translucent folding desk lamp: home office, student dorm, maker bench; each with distinct lighting and readable headlines "Clear Light", "Fold Flat", "Build Bright".

Use case: productivity-visual
Asset type: derived-play contact sheet
Primary request: derive new gameplay from the first group: remix text-heavy poster into recipe content, turn UI into physical scene, turn sequential character into reusable asset sheet, turn product render into campaign variants.
Composition: 2x2 grid, each panel visually distinct but coherent, production-grade, inspectable details.
Avoid: real brands, celebrity likenesses, copyrighted characters, gibberish labels, excessive tiny text, cluttered layout.

Create a polished 2x2 contact sheet titled only with small unobtrusive labels, showing four distinct GPT Image 2 playground use cases. Overall style: premium design research board, clean margins, consistent lighting, 2K-detail look, no brand logos, no watermarks.

Panel A label: "1. Multilingual poster". Visual: a modern street-food poster on a wall, featuring appetizing noodles and small readable bilingual typography in English and Chinese, with the exact visible words "NOODLE NIGHT" and "今晚吃面".
Panel B label: "2. UI snapshot". Visual: realistic desktop app screenshot mockup for a calm personal finance dashboard, charts, transaction rows, readable microcopy like "Monthly flow" and "Subscriptions".
Panel C label: "3. Sequential comic". Visual: four mini comic frames within the panel, same original character: a small red delivery robot preparing tea, leaving a studio, arriving at a desk, and presenting a cup; consistent character across frames.
Panel D label: "4. Product world". Visual: photorealistic hero shot of a translucent modular desk lamp, exploded view hints, material callouts, soft studio shadows, readable tiny label "fold hinge".

Use case: productivity-visual
Asset type: exploratory use-case contact sheet
Primary request: draw a first group of playful GPT Image 2 use cases covering text rendering, UI mockups, sequential consistency, and product visualization.
Composition: 2x2 grid, each panel visually distinct but coherent, high craft, enough detail to inspect the capability.
Avoid: real company brands, celebrity likenesses, copyrighted characters, illegible gibberish text, distorted hands, extra panels.

Create a polished 2x2 contact sheet titled "GPT Image 2: Derived Play Mechanics". Each panel should look like a concrete playable mechanic, not just a poster. Use crisp English labels, clean layout, no brand logos, no watermarks, high-fidelity mockups on real surfaces.

Panel 1 label: "CHOICE ENGINE". Show a recursive story image system: a central generated scene card with exactly three large choice buttons labeled "OPEN", "TRADE", "SING", arrows leading to three new prompt cards. Make it clear that each image becomes the next prompt.

Panel 2 label: "THUMBNAIL LADDER". Show five video thumbnail concepts arranged on a diagonal ladder from "Easy Today" to "Most Clickable". Each thumbnail is about the same topic, a desk coffee gadget, but with different framing and emotion. Include tiny score chips for "shoot" and "click".

Panel 3 label: "DISTRACTOR FORGE". Show an education eval setup: one correct diagram card with a green check and three plausible wrong diagram cards with orange question marks. Topic: simple electric circuit. Include a small strip labeled "Misconception Notes".

Panel 4 label: "PRINT-AND-PLAY BALANCER". Show tabletop card variants in a playtest workspace with visible cost, power, rarity marks, sticky notes saying "too strong", "slow", "fun", and a next-version sheet. It should feel like rapid game balancing.

Overall: four distinct panels, stable grid, sharp readable headings, thoughtful use of color, playful but professional, 16:9 landscape.

Create a polished 2x2 contact sheet titled "GPT Image 2: System Play Use Cases". Each panel should look like a finished concept mockup, with short crisp English labels only. Use a clean editorial design, high fidelity, realistic printed materials mixed with illustration, no brand logos, no watermarks.

Panel 1 label: "BRANCHING PICTURE BOOK". Show an open children's picture book spread with a friendly original character in a cozy workshop, plus exactly three visible illustrated choice cards at the bottom labeled "Door", "Map", "Song". Include tiny next-scene thumbnails connected to those choices.

Panel 2 label: "CREATOR SHOT RECIPE". Show a short-video planning board on a desk: hook frame, prop setup, overhead camera position diagram, caption overlay sample, and thumbnail sketch. It should feel useful for a YouTube/TikTok creator planning a shoot today.

Panel 3 label: "VISUAL EXAM BANK". Show four educational quiz cards with small science diagrams, A/B/C choices, and a discreet answer-key strip. Keep text short and legible: "Force", "Heat", "Orbit", "Signal".

Panel 4 label: "TABLETOP PROTOTYPE KIT". Show a print-and-play board game prototype kit: cards, tokens, player mat, rule reminder, and box-front mockup on a neutral tabletop. Use visible numbers and icons, but keep all text short and clean.

Overall: four distinct panels, consistent margins, sharp readable headings, playful but professional, 16:9 landscape.

Create a second polished 2x3 visual research contact sheet titled "Derived Play Patterns for GPT Image 2.0 / 派生玩法". Aspect ratio 16:9, visually rich, each panel a different playful product/game-like concept built from image generation capabilities. Include clean readable short labels: 1) "REMIX DICE / 变体骰子" showing dice that randomize style, season, camera, and audience for one product image; 2) "VISUAL QUEST / 视觉任务" showing a small quest board where each generated image unlocks the next prompt clue; 3) "BRAND NPC / 品牌角色" showing one mascot character adapted into package, sticker, app avatar, and comic; 4) "MEMORY PALACE / 记忆宫殿" showing a study room where facts become surreal visual landmarks; 5) "A/B ARENA / 视觉竞技场" showing four ad thumbnails competing with score badges like clarity, novelty, click intent; 6) "WORLD KIT / 世界观素材包" showing map, props, character sheet, item cards, and poster in one coherent fictional world. Use original fictional content only, no real brands, no logos, no watermark. Emphasize structured layouts, consistent characters/assets, legible bilingual headings, and playful creative direction.

Create a polished 2x3 visual research contact sheet showcasing six playful GPT Image 2.0 use cases. Aspect ratio 16:9, crisp editorial design, each panel visually distinct, professional but fun. Include short bilingual panel labels in clean readable text: 1) "AD REMIX / 广告变体" showing a sparkling canned drink ad with alternate seasonal props; 2) "UI SNAPSHOT / 界面快照" showing a realistic mobile app screen mockup for a tiny travel planner; 3) "INFOCARD / 知识卡片" showing a clear mini infographic about coffee brewing variables; 4) "COMIC BEAT / 漫画分镜" showing three consistent character moments in one panel; 5) "PRODUCT KIT / 产品套装" showing packaging, sticker, icon, and social post assets; 6) "TEXT LOCALIZE / 文案本地化" showing a poster where English and Chinese text blocks are aligned cleanly. Emphasize accurate typography, structured layout, consistent art direction within each mini-panel, no watermarks, no logos, no real brand names.

Use case: productivity-visual / image evals.
Asset type: valid non-repeating derived-play board for GPT Image 2 research.
Primary request: Create a 16:9 board titled "版式玩法派生" showing four playable derivatives from exact-layout image generation. Panel 01: 约束抽卡 - cards for 日期, 数量, 路线, 预算, 字数. Panel 02: 错误猎人 - a checklist for catching wrong counts, broken grid alignment, bad Chinese text, missing icon, wrong total. Panel 03: 多尺寸改编 - same fictional itinerary converted into square cover, vertical story, thumbnail, and print poster frames. Panel 04: 二次编辑挑战 - before/after slots for changing only weather, changing only price, changing only route, changing only deadline. Chinese labels must be clear and correctly written.
Composition: practical game-like test board for creators, clean grids, small scoring tokens, distinct panel colors, readable typography, no brand marks.
Avoid: repeating product packaging, storyboards, world-bible cards, camera lottery, memory palace, asset reskin dice, food ads, tea infographics, storefront grids, fantasy character art, watermark.

Use case: productivity-visual / infographic-diagram.
Asset type: valid non-repeating use-case board for GPT Image 2 research.
Primary request: Create a 16:9 contact-sheet board titled "精确版式玩法" with four distinct mini experiments, each in its own clean panel: 01 AI旅行日历 with a month grid and tiny itinerary icons; 02 魔法收据 with item rows, prices, subtotal, and a playful fictional receipt layout; 03 城市寻宝地图 with streets, numbered stops, and a legend; 04 会议清单海报 with checkboxes, owners, and deadlines. Chinese labels must be clear and correctly written. The board should show how image generation can create structured artifacts that test layout fidelity, counting, grids, and readable text.
Composition: practical creator lab contact sheet, white background, black grid, restrained accent colors, polished but utilitarian.
Avoid: fantasy characters, food ads, tea steps, hardware product boards, paper boat storyboards, app UI screens, storefront style grids, world-bible cards, unreadable text, brand logos, watermark.

Use case: infographic-diagram / image evals.
Asset type: derived workflow concept image for GPT Image 2 use-case research.
Primary request: Draw a 16:9 quality-control dashboard poster titled "图像玩法评分板". Show six evaluation lanes for generated image experiments: 指令跟随, 文字正确, 风格一致, 局部编辑, 商用可用, 惊喜程度. Each lane has a 0-5 score meter, a tiny pass/fail chip, and a thumbnail placeholder. Text must be clear Chinese. It should feel like a practical visual QA tool for creators testing GPT Image 2 workflows.
Composition: modern dashboard interface mixed with printed lab sheet, readable labels, neutral background, subtle accent colors.
Avoid: gibberish text, brand logos, watermark, overdecorated sci-fi UI.

Use case: illustration-story / compositing concept.
Asset type: derived gameplay concept image for GPT Image 2 use-case research.
Primary request: Draw a 16:9 "world bible generator" visual: one original landscape image in the center is exploded into a coherent story asset sheet around it. Include panels for 主角, 反派, 交通工具, 地标, 气候, 物品, 色板, 三幕剧情. Chinese labels must be clear and correct. The assets should share one consistent whimsical adventure style and look like they belong to the same fictional world.
Composition: professional concept art sheet, clean labeled panels, consistent palette, high detail.
Avoid: inconsistent styles, unreadable labels, brand logos, watermark, too many random objects.

Use case: scientific-educational / illustration-story.
Asset type: derived gameplay concept image for GPT Image 2 use-case research.
Primary request: Draw a 16:9 tabletop game instruction board titled "Prompt Quest 规则卡". It explains a mini game where players roll prompt dice and place cards to build an image prompt. Include four clear Chinese rule blocks with exact labels: 1 抽角色, 2 摆场景, 3 加限制, 4 生成变体. Show dice, tokens, and a sample quest path. Make it visually fun but clean enough to use as a real rule card.
Composition: flat-lay board game rule sheet, readable Chinese headings, charming illustrated components, strong hierarchy.
Avoid: gibberish, too much tiny body text, brand logos, watermark, messy fantasy battle scene.

Use case: product-mockup / ads-marketing.
Asset type: exploratory demo image for GPT Image 2 use-case research.
Primary request: Draw a 3:2 product mockup scene for a fictional creator kit called "Prompt Bento". Show a small boxed kit containing prompt dice, mini storyboard cards, color swatches, and a tiny foldout scene map. The packaging has the exact readable Chinese subtitle "灵感便当" and small English title "Prompt Bento". Make it feel like a premium stationery product photographed for an online shop.
Composition: front-facing three-quarter product photo on a neutral studio table, soft shadows, crisp packaging details, tasteful mixed colors.
Avoid: real brand logos, misspelled title/subtitle, clutter, watermark.

Use case: infographic-diagram.
Asset type: exploratory demo image for GPT Image 2 use-case research.
Primary request: Draw a clean 16:9 Chinese infographic poster titled "一张图变成十种资产". Show a central generated image thumbnail branching into ten neat output tiles: 海报, 封面, 分镜, 贴纸, 商品图, 表情包, UI图标, 教程图, 地图, 卡牌. Make the Chinese text large, clear, and correct. Use a modern editorial infographic layout with subtle grid lines, small arrows, and distinct colors per branch.
Composition: crisp presentation-slide quality, high contrast, readable at thumbnail size.
Avoid: tiny unreadable text, misspelled Chinese, fake brand marks, clutter, watermark.

Use case: productivity-visual / ui-mockup.
Asset type: exploratory demo image for GPT Image 2 use-case research.
Primary request: Draw a polished 16:9 concept image showing a "visual prompt card laboratory" for creators: a tabletop layout with six prompt cards, each card has a tiny image thumbnail, a short Chinese label, and a simple score chip. Labels must be legible: 角色, 场景, 道具, 风格, 约束, 变体. The image should feel like a real photographed design workshop, with paper texture, pens, sticky notes, and a laptop edge visible, but no brand logos.
Composition: top-down editorial product photo, organized but playful, warm natural light, high detail.
Avoid: gibberish text, extra labels, logos, watermarks, fantasy characters.

Create a polished 2x3 contact sheet of six NEW derived playful GPT Image 2 workflows, avoiding prior themes: bilingual food ad, tea infographic, product-board, paper boat storyboard, mood menu, storefront style grid, generic poster/UI/product/infographic/storyboard/interior boards. These should be repeatable game-like mechanisms. No brand logos, no real people identities, no copyrighted character styles. Small clean English labels only: 1) MEMORY PALACE: one abstract concept becomes a navigable illustrated room with mnemonic objects; 2) ASSET RESKIN DICE: a single game icon reskinned by rolling material, era, weather, and rarity dice; 3) CAMERA LOTTERY: the same miniature scene rendered through six camera rules: macro, CCTV, drone, fisheye, x-ray cutaway, thermal; 4) ERROR BEAUTIFIER: broken screenshots, bad scans, and ugly forms transformed into elegant visual systems; 5) WORLD BIBLE CARDS: creatures, maps, tools, laws, and symbols from one fictional culture shown as consistent reference cards; 6) ONE-PROMPT SEQUENCER: one seed phrase unfolds into six adjacent visual states: seed, conflict, transformation, tool, ritual, souvenir. Style: playful design-system laboratory, crisp panel borders, varied palettes, high detail, 16:9 canvas, legible short labels, no watermarks.

Create a polished 2x3 contact sheet of six NEW GPT Image 2 use-case concepts that explicitly avoid these prior themes: bilingual food ads, tea brewing infographics, handheld product boards, paper boat storyboards, mood-menu apps, storefront style grids, generic product shots, generic dashboards, generic eclipse infographics. No brand logos, no real people identities, no copyrighted character styles. Small clean English labels only: 1) RESTORE KIT: an old damaged family photo beside a repaired colorized version, with dust/scratch removal visible; 2) TRY-ON CATALOG: the same neutral mannequin wearing six jacket silhouettes and fabrics in a clean fashion grid; 3) GAME ASSET SHEET: a top-down cozy strategy-game tile set with terrain, buildings, icons, and props; 4) COMIC LOCALIZE: one comic page layout with blank-safe speech bubbles and localized short phrases, showing typography control; 5) FILM PREVIS: a director's shot-planning board showing lens, light, and blocking variations for one scene; 6) DATA POSTER: a beautiful nonliteral poster turning abstract numeric patterns into visual art without charts. Style: premium creative lab presentation, varied color palettes, high detail, crisp panel dividers, 16:9 canvas, professional image-generation showcase, no watermarks.

Create a polished 2x3 contact sheet of six playful derived GPT Image 2 workflows, building on the previous use-case categories but making them more game-like and experimental. No brand logos, no real celebrities, no copyrighted character styles. Use small clean English labels only: 1) PROMPT SLOT MACHINE: three vertical reels showing Subject, Style, Constraint combinations turning into a quirky final image; 2) TEXT STRESS TEST: a wall of tiny posters, each testing short readable phrases, dates, prices, and multilingual snippets without clutter; 3) CONSISTENCY QUEST: the same original character shown across six outfits, moods, and camera angles while preserving identity; 4) EDIT ESCAPE ROOM: one messy room image with highlighted edit missions: remove, replace, relight, recolor; 5) PRODUCT MUTATION LAB: one simple object evolving through six packaging/material/audience variants; 6) MEME TO MOCKUP: a rough napkin sketch transforming into a clean app screen and then a billboard concept. Style: playful lab notebook plus premium design board, high detail, clear panel separation, varied palettes, 16:9 canvas, crisp visual hierarchy, no watermarks.

Create a polished 2x3 contact sheet of six distinct GPT Image 2 use-case concepts. No brand logos, no real celebrities, no copyrighted character styles. Each panel should look like a finished mini artwork, with small clean English labels only: 1) TYPE POSTER: a colorful event poster with crisp readable typography and layered graphic design; 2) UI SCREEN: a realistic SaaS dashboard screenshot mockup on a laptop, dense tables and charts; 3) PRODUCT SHOT: a premium reusable bottle packaging concept with label, box, and studio lighting; 4) INFOGRAPHIC: a clean educational science infographic about how a solar eclipse works, diagram arrows and short labels; 5) STORYBOARD: a four-frame cinematic micro story shown inside one panel, consistent protagonist and environment; 6) ROOM REDESIGN: before-and-after interior design visualization in a small apartment. Style: contemporary editorial design, high detail, varied palettes across panels, 16:9 canvas, sharp composition, professional image generation showcase.

Create a single high-resolution 3-panel derived-play concept board. Three vertical panels side by side, separated by thin grid lines. These are playful derivations from the previous three concepts, not the same images.

Panel 1: Poster-to-AR installation. The "月光档案馆 / MOONLIGHT ARCHIVE" event poster has become an immersive street AR wayfinding scene. A person holds a phone in front of the poster, and the phone screen shows glowing moonlit archive shelves extending into the real street with elegant arrows and tiny floating labels. Keep the bilingual event identity visible but not dominant. Cinematic night photography, plausible UI overlay, legible key words only.

Panel 2: Cutaway-to-learning-kit. The ramen vending machine cutaway has been transformed into a tactile STEM learning kit on a classroom table: magnetic machine layers, labeled cards, tiny pump/valve tokens, a fold-out instruction mat, and kids' hands assembling it. Labels on cards in Chinese: "汤泵", "面仓", "蒸汽阀", "调味轮". Warm documentary product photography, organized parts, educational but fun.

Panel 3: Character-to-merch system. The Lumo moon courier mascot becomes a cohesive merchandise launch board: sticker sheet, enamel pin, tote bag, small plush, postcard, and shipping-box tape. Maintain the same moon-shaped courier with blue scarf and satchel. Include small clean brand text "Lumo Courier" on packaging only. Bright studio flat lay, consistent brand colors, realistic materials.

Overall: polished concept research board, high detail, no watermark, no clutter, keep all text minimal and as specified, avoid copyrighted styles or characters.

Create a single high-resolution 3-panel concept board titled visually by layout only, no large title text. Three vertical panels side by side, each panel is a distinct GPT image generation use-case probe, polished and presentation-ready.

Panel 1: Multilingual localization visual system. A realistic museum night event poster mockup on a wall, with crisp bilingual typography in Chinese and English. Exact visible text: "月光档案馆" and "MOONLIGHT ARCHIVE" plus smaller lines "夜间开放 21:00-01:00" and "Night Hours 21:00-01:00". Sophisticated editorial design, deep ink black, warm paper, subtle silver foil, legible text.

Panel 2: Educational cutaway infographic. A charming but technically clear cutaway illustration of a fictional desktop ramen vending machine, with labeled parts in Chinese: "汤泵", "面仓", "蒸汽阀", "调味轮", "取餐口". The labels should be clean, printed, and connected with thin callout lines. Bright workshop lighting, semi-realistic product illustration.

Panel 3: Character consistency sheet. A small original mascot named "Lumo" (a moon-shaped courier with tiny satchel and blue scarf), shown in six poses on a clean design sheet: neutral, running, waving, carrying parcel, surprised, sleeping. Keep the same character identity, proportions, scarf, and satchel across all poses. No copyrighted characters, no artist imitation.

Overall: each panel separated by thin grid lines, modern design research board, sharp details, strong composition, no watermark, no messy pseudo text beyond the specified words.

Use case: style-transfer
Asset type: playful derived workflow concept, 16:9 landscape
Primary request: Create a "local legend generator" image: one ordinary neighborhood convenience store transformed into four coherent mythic visual styles while keeping the same storefront structure.
Scene/backdrop: a small urban convenience store on a corner, seen from the same front angle in all panels.
Subject: four panels of the same store: 1 cozy watercolor, 2 retro sci-fi neon, 3 ancient ink-painting scroll, 4 clay miniature diorama.
Text: no words except one simple sign that says "OPEN" on each storefront.
Composition: 2x2 grid, identical camera angle and building silhouette across styles, strong style contrast but same core identity.
Style: high craft visual experiment, consistent object identity, fun for social media carousel.
Avoid: brand logos, extra signs, illegible text, chaotic layout, watermark.

Use case: ui-mockup
Asset type: playful derived workflow concept, 16:9 landscape
Primary request: Visualize a "mood-to-menu generator" app screen where a user turns a feeling into a dinner menu and matching table styling.
Scene/backdrop: realistic tablet on a kitchen counter, app UI shown on screen, generated food and table settings around it.
Subject: app interface with three tabs: "Mood", "Menu", "Table". Mood selected: "rainy cozy". Menu cards show "ginger congee", "charred greens", "osmanthus soda".
Text: preserve those exact English labels only, plus a small Chinese title 「情绪菜单」.
Composition: clean believable app mockup, food props around tablet echoing the selected mood, clear screen hierarchy.
Style: sophisticated product/UI photography, warm kitchen lighting, practical not fantasy.
Avoid: fake app store badges, illegible paragraphs, too many screens, watermark.

Use case: illustration-story
Asset type: storyboard contact sheet, 16:9 landscape
Primary request: Create a six-frame silent micro-comic about a tiny paper boat traveling through a rainy city gutter and becoming a lantern reflection at night.
Scene/backdrop: cinematic rainy city street at ground level, puddles, curb, tiny reflections.
Subject: the same folded paper boat across all six frames; preserve identity, proportions, and color across frames.
Text: no words, no captions, no speech bubbles.
Composition: six equal panels, left-to-right progression, clear story beat in every panel, consistent lighting evolution from afternoon rain to night glow.
Style: charming editorial illustration with cinematic lighting, rich details, coherent sequence.
Avoid: text, frame numbers, random characters, inconsistent boat design, watermark.

Use case: product-mockup
Asset type: product concept sheet, 16:9 landscape
Primary request: Create a compact product-visualization board for a fictional portable creative tool called "Pocket Studio".
Scene/backdrop: bright tabletop studio with three realistic product angles.
Subject: a small handheld matte white device with an e-ink screen, side dial, USB-C port, and modular color shells; show front, side, and in-use angle.
Text: include exactly these labels: "Pocket Studio", "Shell swap", "Prompt dial", "USB-C".
Composition: catalog mockup layout with callout lines, color swatches, subtle shadows, very clean spacing.
Style: premium product render, realistic materials, editorial industrial design board.
Avoid: real brand logos, unreadable labels, excessive futuristic shapes, watermark.

Use case: scientific-educational
Asset type: classroom explainer sheet, 4:3 landscape
Primary request: Create a clear educational diagram that demonstrates GPT Image 2 style structured visual reasoning: a four-step illustrated guide to brewing gongfu tea.
Scene/backdrop: clean off-white paper background with subtle grid, infographic layout.
Subject: four numbered panels showing: 1 warm vessel, 2 add leaves, 3 short steep, 4 pour and share.
Text: include exactly these labels in simplified Chinese and English: "1 温杯 / Warm", "2 投茶 / Leaves", "3 冲泡 / Steep", "4 分茶 / Share".
Composition: four equal panels, arrows between panels, simple accurate teaware, legible text, neat icon-like illustrations.
Style: premium educational poster, crisp linework with soft watercolor accents, calm but not monochrome.
Avoid: dense paragraphs, misspelled text, extra steps, watermark, ornate decorations.

Use case: ads-marketing
Asset type: exploratory use-case card, 16:9 landscape
Primary request: Create a polished bilingual street-food campaign poster that demonstrates accurate Chinese and English text rendering inside the image.
Scene/backdrop: bright modern night market kiosk with appetizing steam and clean signage.
Subject: a premium scallion pancake wrap and iced jasmine tea on a counter, photographed like a high-end food ad.
Text: include exactly these visible headline words: 「今晚吃点脆的」 and "CRISPY NIGHT BITES". Include small menu labels: 「葱油饼卷」, "Jasmine Tea".
Composition: poster-ready, centered food hero, enough breathing room, crisp typography, realistic lighting, no fake brand logos.
Style: high-end commercial food photography mixed with clean graphic design, professional print ad, 2K-feeling detail.
Avoid: misspelled text, unreadable tiny text, watermark, extra slogans, distorted hands.

Use case: ads-marketing / derived play mechanic: product-to-flavor-lab
Asset type: triptych campaign image
Primary request: Derive a playful flavor lab from the "Cloud Oolong" can: show three experimental flavor variants as a polished campaign lineup.
Exact visible text on cans: "Cloud Oolong", variant names "Jasmine Peel", "Lychee Rain", "Roasted Pear".
Scene/backdrop: bright studio product lineup, three cans centered, each with matching ingredient vignette and color accent.
Subject: consistent packaging system across all three cans, condensation, subtle reflections, premium sparkling tea mood.
Style: high-end beverage campaign triptych, clean typography, realistic materials, no real brand resemblance, no watermark.

Use case: ui-mockup / derived play mechanic: app-to-board-game
Asset type: tabletop board game prototype photo
Primary request: Transform the "Route Bento" travel planner concept into a physical board game called "Route Bento: City Dash".
Exact visible text: box title "Route Bento: City Dash", card labels "Map", "Budget", "Food", "Walk".
Scene/backdrop: overhead tabletop photo, game board with city grid route, small tokens, cards, budget tracker, food tiles.
Subject: a complete playable-looking board game kit inspired by itinerary planning.
Style: crisp product prototype photography, clear graphic design, readable labels, no real brands, no watermark.

Use case: illustration-story / derived play mechanic: poster-to-mini-comic
Asset type: four-panel visual story page
Primary request: Derive a playful image-generation game from the "midnight bookstore" poster: a four-panel mini-comic where one old book opens a tiny night city on the reading table.
Exact visible text: small title at top "Book Portal", panel labels "1", "2", "3", "4" only.
Scene/backdrop: same cozy rainy bookstore mood, but more whimsical and sequential.
Subject: open book gradually becoming a miniature glowing city, with tiny streets and lights rising from the pages.
Style: polished graphic novel page, consistent lighting across panels, legible panel structure, no real characters, no watermark.

Use case: product-mockup / imaginary packaging system
Asset type: product photography concept
Primary request: Create a premium product mockup for an imaginary sparkling tea drink named "Cloud Oolong".
Exact visible text on can: "Cloud Oolong", "Sparkling Tea", "0 Sugar", "Jasmine Peel".
Scene/backdrop: bright tabletop product photo with condensation, one can upright, one can lying sideways, citrus peel and tea leaves around it.
Subject: elegant aluminum can packaging with legible label typography and cohesive visual system.
Style: commercial product photography, high realism, clean lighting, fresh palette, no real brand resemblance, no watermark.

Use case: ui-mockup / layout-aware screenshot
Asset type: clean mobile app mockup image
Primary request: Create a realistic phone screenshot mockup for an imaginary app called "Route Bento", a travel micro-itinerary planner.
Exact visible UI labels: "Route Bento", "Today", "Map", "Budget", "Food", "2h walk", "$38 left".
Scene/backdrop: neutral desk surface, phone slightly angled, soft daylight, no hands.
Subject: a dense but polished mobile interface with a map card, timeline stops, budget chip, and food recommendations.
Style: modern product design screenshot, crisp typography, restrained colors with one accent, no real brand icons, no watermark.

Use case: ads-marketing / typography stress test
Asset type: square social poster concept, Chinese-first visual
Primary request: Create a playful, production-polished poster for an imaginary late-night bookstore event.
Exact visible text: main title "午夜书店", subtitle "今晚 9:30", small footer "只带一本旧书入场". Render the Chinese text cleanly and legibly.
Scene/backdrop: cozy urban bookstore at night, warm window light, rain-speckled glass, shelves receding into depth.
Subject: a glowing reading table with three open books, gentle cinematic atmosphere.
Style: editorial poster, tasteful typography hierarchy, no real brands, no logos, no watermark.

Use case: ui-mockup
Asset type: in-product desktop web app screen
Primary request: create a corrected realistic software UI showing parent/current/child three-layer node navigation with the proper card size order. Parent is the largest base card, current is a smaller middle card above it, and child is the smallest foreground detail card. This nested size hierarchy must be visually clear.
Scene/backdrop: straight-on desktop browser screenshot of a calm reading workspace with sidebar and warm editorial app styling.
Subject: left sidebar tree navigation and central stack of three cards. The parent card is the largest contextual layer in the back, the current reading card is medium and centered on top of it, and the child drill-down card is the smallest panel floating at front-right.
Style/medium: production-grade web app interface, believable and usable, warm literary SaaS design.
Composition/framing: direct desktop screenshot, enough zoom to clearly read the size progression from parent to current to child.
Lighting/mood: soft warm daylight, quiet, premium, thoughtful.
Color palette: oat cream, ivory, sand, terracotta, charcoal, with only very subtle muted sage as optional secondary accent.
Materials/textures: paper-like surfaces, refined borders, soft layered shadows, restrained overlay effects.
Text (verbatim): small labels allowed for Parent, Current, Child.
Constraints: parent card must be largest, current smaller, child smallest; light mode only; no dark mode; no purple; no neon; no generic dashboard feel; keep the hierarchy tactile and calm; match the current website's warm editorial theme.

Use case: ui-mockup
Asset type: landing page hero showcase for a reading and knowledge app
Primary request: create a corrected marketing UI design showing a three-layer node concept with proper physical card hierarchy. The parent card is the largest card in the stack, the current card is a medium card centered above it, and the child card is the smallest floating foreground card. The size relationship must be obvious immediately.
Scene/backdrop: warm editorial landing page section on a soft oat-cream background with faint paper grain and restrained radial light.
Subject: three nested content cards representing parent, current, child. Parent is the broadest and tallest background card with overview structure. Current is a smaller middle card with focused reading content. Child is the smallest compact detail card hovering in front.
Style/medium: premium SaaS product mockup, refined editorial interface, believable Figma-quality composition.
Composition/framing: desktop hero composition, left side brief value proposition text, right side stacked card mockup. Parent visibly spans the widest area, current is clearly inset, child is clearly the smallest accent card.
Lighting/mood: soft daylight, warm, calm, intelligent.
Color palette: bone, cream, sand, terracotta, muted amber, charcoal.
Materials/textures: paper-like cards, thin warm borders, soft shadows, subtle terracotta highlight around the smallest child card only.
Text (verbatim): tiny labels allowed: Parent, Current, Child.
Constraints: parent must be largest card, current medium, child smallest; light mode only; no purple; no dark mode; no glassmorphism; no neon; no clutter; keep the current website's warm literary style.

Use case: ui-mockup
Asset type: landing page hero showcase for a reading and knowledge app
Primary request: redesign the parent/current/child three-layer node concept with the CORRECT size hierarchy. The parent card must be the largest background layer, the current card must be a medium-sized card placed above it, and the child card must be the smallest foreground card. All three should read as nested cards getting smaller as depth increases.
Scene/backdrop: airy light-theme landing page section on a warm oat-cream background with subtle paper texture and quiet radial light.
Subject: a large centered product mockup showing three stacked cards. Parent is a wide, dominant background card with broad context content. Current is a clearly smaller middle card centered in front of the parent. Child is the smallest floating detail card in the foreground, positioned slightly to the right.
Style/medium: premium product design mockup, refined editorial interface, calm literary feeling, high-end Figma/Dribbble presentation quality.
Composition/framing: desktop website hero composition, generous whitespace, asymmetric but balanced layout, strong visual focus on the card-size hierarchy. The difference in size between parent, current, and child should be obvious at first glance.
Lighting/mood: soft daylight, warm and intelligent, premium but restrained.
Color palette: warm cream, bone, sand, terracotta, muted amber, charcoal text.
Materials/textures: paper-like cards, thin warm-gray borders, soft layered shadows, subtle terracotta glow around the smallest child layer.
Text (verbatim): tiny UI labels allowed for Parent, Current, Child, but keep text minimal.
Constraints: light theme only; elegant serif display feel mixed with clean sans-serif UI details; no dark mode; no purple; no neon blue; no cyberpunk; no glassmorphism; no stock-photo people; avoid clutter; parent must be visually largest, current medium, child smallest.

Use case: ui-mockup
Asset type: in-product desktop web app screen
Primary request: create a realistic software UI showing a reading app with parent/current/child three-layer node navigation in actual use.
Scene/backdrop: desktop browser screenshot of a calm knowledge-reading workspace.
Subject: left sidebar tree navigation, central reading canvas, and a layered node interaction. The parent card is tucked behind and peeking from the top, the current card is the main readable article surface, and the child node is opened as a modal sheet hovering above the current card with a slightly stronger accent and backdrop blur.
Style/medium: production-grade web app interface, modern editorial product UI, believable and usable, not concept art.
Composition/framing: straight-on desktop screenshot, slightly zoomed in so the three layers are obvious and tactile.
Lighting/mood: soft warm daylight, thoughtful, quiet, premium.
Color palette: oat cream, soft ivory, terracotta, sand, charcoal, tiny muted sage accents only if subtle.
Materials/textures: paper-like surfaces, rounded cards, fine borders, soft shadows, subtle modal overlay, restrained browser chrome.
Text (verbatim): minimal readable UI text only, mostly short headings and paragraph blocks, avoid lots of typography.
Constraints: reference a warm editorial website theme rather than a generic dashboard; light mode only; no dark mode; no purple; no neon; no glossy glass cards; no excessive charts; keep the hierarchy readable and calm; make the parent/current/child depth relationship the main design idea.

Use case: ui-mockup
Asset type: landing page hero showcase for a reading and knowledge app
Primary request: create a polished marketing UI design that demonstrates a parent/current/child three-layer node concept, inspired by a warm editorial SaaS brand.
Scene/backdrop: airy light-theme landing page section on a warm oat-cream background with subtle paper texture and quiet radial light.
Subject: a large centered product mockup that clearly shows three depth levels: a faded parent card peeking from behind at the top, a crisp current reading card in the middle, and a floating child detail card above and slightly to the right.
Style/medium: premium product design mockup, refined editorial interface, calm literary feeling, high-end Figma/Dribbble presentation quality.
Composition/framing: desktop website hero composition, generous whitespace, asymmetric but balanced layout, strong visual focus on the stacked card hierarchy.
Lighting/mood: soft daylight, warm and intelligent, premium but restrained.
Color palette: warm cream, bone, sand, terracotta, muted amber, charcoal text.
Materials/textures: paper-like cards, thin warm-gray borders, soft layered shadows, subtle terracotta glow around the active child layer.
Text (verbatim): keep text minimal, mostly abstract UI lines and a few tiny labels only.
Constraints: light theme only; elegant serif display feel mixed with clean sans-serif UI details; no dark mode; no purple; no neon blue; no cyberpunk; no glassmorphism; no stock-photo people; avoid clutter; emphasize the three-layer parent/current/child feeling.

Use case: infographic-diagram
Asset type: Drill feature marketing visual, wide 16:9 section image
Primary request: Create a clean minimal product illustration that explains the three-step Drill workflow: highlight a confusing phrase, open an inline explanation, then drill one level deeper while keeping a visible path back.
Scene/backdrop: light editorial canvas with structured UI cards and layered reading panels, no background clutter.
Subject: a simple elegant visualization of recursive reading comprehension, with one paragraph panel, one explanation panel, and one deeper nested panel connected clearly.
Style/medium: refined modern infographic illustration, product-design aesthetic, minimal but premium, suitable for SaaS feature sections.
Composition/framing: wide horizontal layout, balanced left-to-right flow, ample whitespace, crisp hierarchy, easy to crop.
Lighting/mood: bright, precise, calm, trustworthy.
Color palette: soft ivory, graphite, cyan, teal, restrained amber for emphasis.
Materials/textures: flat-glass cards, subtle shadows, thin connector lines, very clean surfaces.
Constraints: no readable text, no logos, no watermark, no generic stock icons, no clutter, visually communicate highlight -> explain -> drill deeper -> zoom back.

Use case: photorealistic-natural
Asset type: Drill marketing campaign image, wide 16:9 visual for landing page or social ad
Primary request: Create a realistic learning scene showing someone reading a dense research paper or technical article on a laptop or tablet, using Drill to understand a highlighted phrase inline without leaving the page.
Scene/backdrop: focused desk setup in natural daylight, modern but not flashy, with a screen showing a clean reading interface and subtle inline explanation panels emerging near highlighted text.
Subject: a thoughtful knowledge worker or student in the middle of understanding a difficult passage, with the product experience centered on in-context comprehension rather than chat.
Style/medium: premium editorial product photography, realistic, crisp, believable SaaS marketing image.
Composition/framing: wide composition with the user and device on the left or center-left, generous clean negative space on the right for copy overlay, screen visible but not overloaded.
Lighting/mood: calm, smart, productive, high-clarity morning light.
Color palette: warm neutrals, paper white, charcoal, muted cyan interface accents.
Materials/textures: laptop aluminum, paper notes, subtle screen glow, realistic skin and textile textures.
Constraints: no readable UI text, no logos, no watermark, avoid stock-photo cheesiness, avoid exaggerated futuristic holograms, keep the interface grounded and product-like.

Use case: stylized-concept
Asset type: Drill website landing page hero image, wide 16:9 marketing visual
Primary request: Create a bold abstract brand visual for Drill, an AI reading comprehension product that lets users drill into anything they read.
Scene/backdrop: a dark-ink editorial page or dense paragraph dissolving into a luminous branching knowledge tree, with subtle interface layers and depth cues emerging from highlighted phrases.
Subject: the transformation from confusing text into structured understanding, shown as glowing recursive layers expanding out of a single highlighted sentence.
Style/medium: premium product-brand illustration, cinematic 3D-meets-editorial, polished startup marketing aesthetic, distinctive not generic, high contrast, refined detail.
Composition/framing: wide composition with strong negative space on the right for headline and CTA overlay, focal energy on the left-center, clean readable silhouette.
Lighting/mood: intelligent, exploratory, precise, calm but powerful; soft volumetric glow from the knowledge tree.
Color palette: off-white paper tones, deep graphite, electric cyan, subtle teal, restrained warm highlights.
Materials/textures: paper grain, glassy UI layers, soft light bloom, crisp typographic-texture hints without readable text.
Constraints: no readable text, no logos, no watermark, no clutter, no purple-dominant palette, suitable for SaaS marketing hero use.