Most AI image prompt libraries are written for a model that does not exist — a hypothetical generalist that renders text well, follows layouts literally, and holds characters consistent. Before GPT Image 2 shipped on 2026-04-21, those libraries were aspirational. With GPT Image 2 now live on Oakgen, they describe actual output.
Below are 50 prompts in five categories we have been running against GPT Image 2 for the past week. Each prompt is copy-paste ready and tuned to the model's strengths — typography, structural layouts, character consistency, multilingual text. Each carries a one-line tip for what to tweak. The prompts work on the model page at /models/gpt-image-2 or anywhere on Oakgen that lets you select a model.
GPT Image 2 treats prompts more literally than any model before it. Change the specific words, dates, names, fonts, and dimensions — keep the structural scaffolding (orientation, layout language, text-size callouts, palette instructions). The scaffolding is what triggers the model's structural-layout behavior.
1. Typographic Posters (10 prompts)
GPT Image 2's single biggest win. If the prompt describes type hierarchy in words, the model obeys.
1. Indie film festival poster.
A2 portrait film festival poster. Title "NORTHBOUND" in 120pt condensed display serif, vertically stacked across left third. Subtitle "A Seven-Film Retrospective, April 12–19" in 16pt italic below. Right half: a grainy black-and-white medium shot of a woman in a peacoat looking past camera, 16mm film grain. Bottom strip: seven director names in 9pt sans caps, pipe-separated. Palette: bone white, charcoal, one drop of burnt orange. Texture: subtle offset-print registration marks.
Tip: swap the title and director names; keep the "120pt / 16pt / 9pt" language — that is what forces the type hierarchy.
2. Jazz night at a small venue.
Jazz night poster, 11x17 portrait. Venue name "The Parlour Room" in 36pt art-deco caps across top, gold foil look on deep navy. Hero line: "Saturday Sessions with Moira Oduya Trio" in 80pt hand-lettered brush script. Date "May 3, 2026 · 9 PM" in 14pt italic. Bottom third: a stylized silhouette of a double bass in single-color linocut. Four small text blocks at the bottom edge: cover charge, dress code, reservation line, door time.
Tip: art-deco + brush script in one poster is a GPT Image 2 showpiece — previous models blur the boundary between fonts.
3. Minimalist lecture poster.
Swiss-style academic lecture poster, A3. Massive headline "On the Limits of Generalization" in 180pt Helvetica, left-aligned, bleeding off the right edge. Below in 18pt: "Dr. Emmanuel Osei · Dept. of Computer Science · Wednesday, May 6, 4:00 PM · Hall 204." Abstract in 9pt justified text, 3 paragraphs, bottom third. Single accent: red bar along the left margin, 6pt wide. Pure white background. No imagery.
Tip: the "3 paragraphs" count is a hard instruction. Try "5 paragraphs" or "2 paragraphs" — the model recounts.
4. Bookstore reading announcement.
Independent bookstore event poster, tabloid portrait. Hand-drawn border of books stacked around the edges, pen-and-ink style. Top: "An Evening With" in 24pt italic serif. Center: author name "ISABELLE PERRAULT" in 100pt serif display caps, slightly condensed. Below: book title "The Cartographer's Apology" in 22pt italic. Date block: "Friday, May 15 · 7 PM · Inkwell Books · 421 Pine Street." Bottom: "Signing and Q&A to follow · Seating limited." Cream paper texture, one-color print (deep forest green).
Tip: the ink-border plus typography combination defeats most image models. GPT Image 2 keeps them separated.
5. Climate rally poster.
Protest poster, A2 portrait, high-contrast screenprint aesthetic. Top half: a stylized woodcut-style illustration of a tree with its roots cracking asphalt. Bottom half: text reads "NO MORE WAITING" in 140pt stencil sans, all caps, ink-splatter edges. Below in 18pt: "Climate March · City Hall · Saturday, June 6 · 11 AM." Palette: kraft paper tan, matte black, one accent red. DIY mimeograph feel.
Tip: ink-splatter edge effects stay within the letterforms — don't bleed into the whole image.
6. Food-hall launch poster.
Grand-opening poster for a food hall, 18x24 portrait. Top: "NOW OPEN" in 90pt condensed caps, vertical. Middle: four rectangular cards arranged in a 2x2 grid, each with a vendor name and cuisine in 20pt: "KIRA · Sichuan noodles", "BASMA · Palestinian mezze", "OAK & EMBER · wood-fired pizza", "TARO · Filipino rice bowls". Bottom: address and hours in 12pt. Palette: warm cream, terracotta, deep olive. Retro 1970s supergraphic style.
Tip: the 2x2 vendor grid is a structural layout — GPT Image 2 preserves the grid cleanly.
7. Opera season announcement.
Opera season poster, portrait orientation. Top banner: "2026–27 SEASON" in 22pt spaced caps. Center: a dramatic oil-painted portrait of a woman mid-aria, stage-lit against black, brushstrokes visible. Below the portrait: five opera titles listed in 16pt italic serif, each with composer and date — "La Traviata / Verdi / Nov 2", "Salome / Strauss / Jan 18", "Akhnaten / Glass / Mar 3", "Eugene Onegin / Tchaikovsky / Apr 14", "Die Zauberflöte / Mozart / Jun 9". Footer: box office phone and website in 10pt. Palette: deep maroon, ivory, gold leaf accents.
Tip: GPT Image 2 will correctly render "ö" and "é" — diacritics were unreliable on earlier models.
8. Skate-shop event.
Skate shop best-trick-contest poster, 11x17. Top: "BEST TRICK 2026" in 150pt graffiti-tag-style display, aggressive. Middle: a stylized illustration of a skateboard mid-kickflip, riso-print two-color aesthetic. Bottom third split into three columns: "WHEN · Sat May 30 · 2 PM", "WHERE · Riverside Bowl", "PRIZE · $500 cash + deck". Palette: neon pink, black, one accent green. DIY zine energy.
Tip: three-column bottom layouts are underused and GPT Image 2 handles them cleanly.
9. Dance performance poster.
Contemporary dance performance poster, A2 portrait. Full-bleed long-exposure photograph of a dancer mid-leap, motion blur trails against pitch black. Text overlaid in crisp white sans-serif: title "BODIES IN MOTION" in 80pt across the upper third. Choreographer "Léa Bonnet" in 22pt italic below. Company name "Ensemble Atrium" in 14pt spaced caps. Tour dates listed in three columns at the bottom: "Paris · Lyon · Bordeaux", each with venue and date. Credits in 8pt along the left edge, rotated 90 degrees.
Tip: "rotated 90 degrees" for the credits is exact — GPT Image 2 respects the rotation.
10. Coffee shop menu board.
Chalkboard-style coffee shop menu, landscape orientation, 36x24. Header: "OAKLAND BREW CO." in 60pt hand-lettered serif caps with chalk-texture shading. Body divided into three columns — "ESPRESSO", "DRIP", "COLD" — each with five drinks and prices in 18pt hand-lettered style. Small chalk illustrations between sections: a portafilter, a pour-over dripper, an ice cube. Subtle smudge texture across the board. White and pastel-chalk colors on deep green chalkboard.
Tip: hand-lettered price consistency across columns is a newer capability — check alignment.
2. UI Mockups (10 prompts)
GPT Image 2 is the first model that can credibly one-shot a design mockup from a prose description.
11. SaaS dashboard hero.
Clean SaaS dashboard UI mockup, 16:9, screenshot framing. Left sidebar: 6 icon+label navigation items ("Dashboard", "Projects", "Analytics", "Users", "Billing", "Settings") with "Dashboard" highlighted. Top bar: logo "LINEAR METRICS" on the left, search field in the center, user avatar + notification bell on the right. Main canvas: 2x2 stat cards at top ("MRR: $42.8K · +12%", "Users: 1,284 · +3.2%", "Churn: 2.1% · -0.4%", "NPS: 67 · +5"). Below: a line chart spanning full width titled "Revenue Trend · Last 90 Days". Right panel: activity feed with 4 recent items. Palette: near-white background, charcoal text, one accent blue. Inter font throughout.
Tip: GPT Image 2 now renders numbers inside chart UIs without butchering them. Verify decimals before shipping.
12. Mobile app login screen.
iOS mobile app login screen, iPhone 17 Pro frame. App name at top in 28pt SF Pro: "Meridian". Subtitle in 14pt gray: "Plan your week, honestly." Center: two stacked input fields labeled "Email" and "Password" with rounded corners, 2pt border. Below: a primary button "Sign In" in solid black with white text. Below that: "Continue with Apple" and "Continue with Google" as outlined secondary buttons. Bottom: "New to Meridian? Create an account" as a centered text link in blue. Status bar: 9:41, full signal, full battery. Realistic iOS pixel density.
Tip: "9:41" is the iOS convention. GPT Image 2 renders status bars correctly now — it did not used to.
13. Settings page with toggles.
Web app settings page mockup, desktop resolution. Page title: "Account Settings" in 32pt bold. Three sections with headers in 18pt spaced caps: "PROFILE", "NOTIFICATIONS", "PRIVACY". Under PROFILE: avatar upload + three text fields ("Name: Maya Chen", "Email: [email protected]", "Timezone: America/Los_Angeles"). Under NOTIFICATIONS: four rows, each with label + iOS-style toggle (three on, one off). Under PRIVACY: three rows, all toggles off. Save button at bottom right. Palette: white, neutral grays, one accent teal for active toggles.
Tip: "three on, one off" is a countable instruction — verify the model honored it.
14. Onboarding multi-step.
Onboarding modal mockup, step 2 of 4. Progress bar at the top showing 50% fill with "Step 2 of 4" label in 12pt. Title in 28pt: "Which tools do you use today?" Subtitle in 14pt gray: "Pick as many as apply. You can change this later." Grid of 8 tool cards (2 rows of 4), each with a logo placeholder and name: "Figma", "Notion", "Linear", "Slack", "Jira", "Miro", "Loom", "GitHub". Three of the cards have a blue border indicating selection. Bottom bar: "Back" (secondary) and "Continue" (primary, blue) buttons.
Tip: "three of the cards have a blue border" — the model counts selections accurately.
15. Pricing page.
SaaS pricing page mockup, desktop width. Top: heading "Simple pricing, scale when ready" in 48pt. Three plan cards side by side. Card 1: "Free · $0/month · 100 generations, community support". Card 2 (featured, outlined in purple, "Most Popular" badge): "Pro · $29/month · Unlimited generations, priority support, team seats". Card 3: "Enterprise · Custom · SSO, audit logs, dedicated CSM". Below: FAQ accordion with 4 collapsed items and the first one ("Can I cancel anytime?") expanded with a short answer. Palette: off-white, dark text, one accent purple.
Tip: pricing cards are where structural layouts really earn their keep — three-card grids used to require post-processing.
16. Calendar view.
Calendar web app mockup, month view, May 2026. Header: "May 2026" in 24pt with prev/next arrows. Below: 7-column grid with day-of-week labels ("Sun Mon Tue Wed Thu Fri Sat"). 5 weeks of cells, current day (May 18) highlighted with a colored ring. Five events placed across the month as colored pills with truncated titles: "Team offsite" (blue, May 4), "Product review" (green, May 8), "Q2 planning" (orange, May 14), "Conference" (red, May 22), "1:1 with Raj" (gray, May 29). Right sidebar: mini month view + "Upcoming" list of 3 events.
Tip: 7-column grids with correctly-aligned day labels are a known weak point of older models; GPT Image 2 fixed it.
17. Email inbox.
Email client web app mockup, two-pane layout. Left pane: inbox list with 8 emails. Each row: sender name, subject line, preview text, timestamp. Three rows are bold (unread). Subjects include "Re: Q2 planning deck", "Shipping update: Order #4821", "Weekly digest · 5 new follows", and "Your invoice is ready". Right pane: open email with full header ("From: Priya Natarajan · To: you · May 18, 3:42 PM · Subject: Re: Q2 planning deck"), body text of 3 short paragraphs, and a reply box at the bottom with "Send" button.
Tip: email previews in 8-12pt type used to come out as gray lorem-ipsum blur. Now they render as real sentences.
18. Analytics chart.
Analytics dashboard card mockup, standalone. Title: "Daily Active Users · Last 30 days" in 18pt bold. Below title: big number "24,813" in 48pt, with "+8.4% vs previous period" in 14pt green. Below the number: an area chart filling the card width, x-axis labeled "May 1" to "May 30" at 5-day intervals, y-axis with four tick labels "0", "5k", "15k", "25k". Smooth curve with a slight upward trend. Palette: white card, charcoal text, translucent teal fill under the curve.
Tip: if the axis labels drift, specify them explicitly as shown.
19. Chat UI.
Chat application UI mockup, mobile portrait. Top bar: contact name "Aiko Tanaka" with online status dot, back arrow, call and video icons. Chat area: 6 alternating message bubbles. Right-side (blue, user): "hey — you free for dinner thursday?", "7ish?", "wear a jacket, it's outdoors". Left-side (gray, them): "sounds good — where?", "see you then 🙂". Each message has a small timestamp below. Input bar at bottom with placeholder "Message…", attachment icon, and send button. iOS pixel-perfect rendering.
Tip: emoji inside bubbles now renders without artifacts.
20. File browser.
File browser UI, grid view, macOS-style. Title: "Design Assets". Breadcrumb: "Projects / Brand Refresh / Design Assets". Grid of 12 file tiles, each showing a file icon and filename: "logo-primary.svg", "logo-secondary.svg", "brandbook-v3.pdf", "color-palette.png", "type-specimen.pdf", "icon-set-32.ai", "icon-set-64.ai", "pattern-tile.png", "cover-hero.jpg", "cover-hero-2.jpg", "moodboard.pdf", "guidelines.pdf". One file is selected (highlighted blue border). Right sidebar: selected file details — name, size, created date, tags.
Tip: filenames with extensions are a great GPT Image 2 torture test — model now gets them right.
3. Infographics (10 prompts)
GPT Image 2 respects label→arrow→target relationships when you specify them explicitly. Use phrasing like "arrow from [source] to [target], labeled [text]" rather than leaving the connective structure implicit.
21. Water cycle.
Educational infographic of the water cycle, landscape orientation. Central illustration: mountain, river, ocean, sun, clouds, forest. Five labeled arrows showing the stages, each arrow with a small caption: "1. Evaporation (ocean → sky)", "2. Condensation (water vapor → clouds)", "3. Precipitation (clouds → mountain)", "4. Runoff (mountain → river)", "5. Collection (river → ocean)". Muted watercolor illustration style with a cream background. Title at top: "The Water Cycle" in 32pt serif.
Tip: numbered stages (1–5) force the arrow ordering.
22. Startup funding stages.
Business infographic: startup funding stages, horizontal timeline. Five stages left-to-right, each with a circular milestone marker and label: "Pre-seed · $50K–$500K · friends, family, angels", "Seed · $500K–$3M · seed funds", "Series A · $3M–$15M · traditional VC", "Series B · $15M–$50M · growth VC", "Series C+ · $50M+ · late-stage funds". Each marker sized proportional to dollar range. Subtle grid background. Palette: navy, off-white, one accent coral for the markers. Title: "Startup Funding Stages at a Glance".
Tip: "sized proportional to dollar range" triggers visual hierarchy — markers scale.
23. Human brain regions.
Anatomical infographic of the human brain, side profile. Clean vector-style illustration of the brain divided into four color-coded regions: frontal lobe (blue), parietal lobe (green), temporal lobe (orange), occipital lobe (purple). Each region labeled with a lead line pointing to it, label includes region name and one function: "Frontal Lobe · decision-making", "Parietal Lobe · spatial reasoning", "Temporal Lobe · language & memory", "Occipital Lobe · vision". Title: "The Four Cerebral Lobes". Palette: muted tones, cream background.
Tip: "lead line pointing to it" is the exact phrase that produces callout lines instead of overlaid text.
24. Carbon footprint breakdown.
Carbon footprint pie chart infographic, portrait orientation. Title: "Where Your Household Carbon Goes" in 28pt bold. Large pie chart with 5 slices labeled with percentage and category: "Transport · 29%", "Home energy · 22%", "Food · 20%", "Goods & services · 18%", "Other · 11%". Each slice in a distinct muted tone. Below the pie: a small "How to reduce" list with one tip per category (5 bullets total). Palette: earth tones on off-white. Source line in 8pt at the bottom: "Source: EPA 2025 household estimate".
Tip: pie-chart percentages must sum to 100 — GPT Image 2 usually gets this right now, but verify.
25. Historical timeline.
Historical timeline infographic: history of photography, 1826 to 2026. Horizontal timeline with 10 evenly spaced events, each with a small illustration and a label: "1826 · First permanent photograph (Niépce)", "1839 · Daguerreotype announced", "1888 · Kodak No. 1 camera", "1907 · Autochrome color process", "1935 · Kodachrome film", "1975 · First digital camera (Kodak prototype)", "1991 · Nikon F3 digital", "2000 · First camera phone (Sharp J-SH04)", "2010 · Instagram launches", "2026 · GPT Image 2 launches". Palette: sepia fading to color across the timeline.
Tip: the sepia-to-color gradient across the timeline is a GPT Image 2 specialty — it used to flatten.
26. Recipe infographic.
Visual recipe infographic: sourdough bread in 6 steps, portrait. Title: "Sourdough Bread · 6 Steps" in 28pt. Six numbered illustrated steps arranged in a 2x3 grid: "1. Mix flour + water + starter", "2. Bulk ferment 4–6h", "3. Shape the dough", "4. Cold proof overnight", "5. Bake in Dutch oven 20m covered", "6. Uncover and bake 20m more". Each step has a small illustration and time indicator. Flour-dusted texture background. Palette: warm cream, rust, black.
Tip: 2x3 grids of illustrated steps are where GPT Image 2's structural layouts shine.
27. Climate change bar chart.
Climate infographic: annual global temperature anomaly, 1880 to 2025. Bar chart with bars for every decade (15 bars total), color-coded from cool blue (1880s, below baseline) through white (mid-20th century baseline) to deep red (2020s, highest anomaly). Y-axis: "Temperature anomaly (°C)" with tick marks at -0.4, 0, +0.4, +0.8, +1.2. X-axis: decade labels from "1880s" to "2020s". Title: "Earth Is Warming". Subtitle: "Decadal averages of global surface temperature anomaly, relative to 1951–1980." Source: "NASA GISS, 2025". Clean editorial style.
Tip: the °C symbol survives across GPT Image 2 — was flaky before.
28. Org chart.
Company org chart infographic, top-down hierarchy. Top: "CEO · Amara Okonkwo". Second row: four direct reports with titles — "CTO · Jin Park", "COO · Liv Andersen", "CMO · Diego Rojas", "CFO · Priya Shah". Under CTO, three boxes: "VP Eng", "VP Product", "VP Design". Under COO, two boxes: "Head of People", "Head of Ops". Under CMO, two boxes: "Growth Lead", "Content Lead". Clean flow-chart style with thin connector lines. Palette: navy boxes on white, one accent gold for executive row.
Tip: hierarchy depth is instructable — "three boxes under CTO" gets respected.
29. Habit tracker.
Wellness infographic: a 30-day habit tracker grid. Title: "30 Days of Morning Walks" in 28pt hand-lettered. Grid of 30 small circle checkboxes arranged in 6 rows of 5, numbered 1–30. First 18 circles filled in with a soft green check; circles 19–30 empty. Below the grid: a motivational line in 14pt italic — "Missing a day is fine. Missing two is a pattern." Palette: cream paper, soft greens, hand-drawn feel.
Tip: "first 18 circles filled" — exact numeric state. The model counts.
30. Personal finance breakdown.
Personal finance infographic: monthly budget breakdown, donut chart. Title: "A Balanced Monthly Budget" in 28pt. Central donut chart with 6 slices, each labeled with category and percentage: "Housing 30%", "Food 15%", "Transport 10%", "Savings 20%", "Debt 10%", "Other 15%". Below the donut: 6 tip cards, one per category, each with a one-line rule ("Housing: aim for 30% or less of take-home" etc.). Palette: teal, gold, cream. Clean financial-editorial style.
Tip: the 6 tip cards matching the 6 slices is the structural-layout payoff — previously these would desync.
4. Manga & Comics Panels (10 prompts)
31. Four-panel manga.
A 4-panel manga page, vertical arrangement, black ink on cream paper. Panel 1: wide establishing shot of a high school rooftop at sunset, a single student leaning against the railing. Panel 2: close-up on the student's face, eyes wide, sweat drop, surprised. Panel 3: reverse angle showing a second student approaching with a wrapped bento box. Panel 4: both students sitting side by side, bento open between them, warm light. Speech bubbles in English: Panel 2: "H-huh?!", Panel 3: "Made extra.", Panel 4: "...thanks." Consistent character designs across all panels. Screentone shading, visible ink hatching.
Tip: "consistent character designs across all panels" is the magic phrase for 8-image coherence within a multi-panel composition.
32. Action splash page.
Comic book splash page, single full-bleed panel, dramatic angle. A cyberpunk courier mid-sprint across a rain-slick Tokyo rooftop, neon signs bleeding color across the wet concrete. Motion lines from left, perspective tilted. Sound effect "KRAK-OOM!" in bold red display lettering across the upper right. Caption box at bottom: "23 seconds since the courier left the Mori Building." Speech balloon tail from the courier's mouth: "Just get me to the bridge." Full-color comic art, heavy ink lines, cinematic lighting.
Tip: sound-effect typography inside the art is a GPT Image 2 exclusive — earlier models rendered "KROOM!" as decoration.
33. Shojo-style close-up.
Shojo manga close-up panel, portrait orientation. A girl with long dark hair and soft features looking down, a single tear forming at the corner of her eye. Soft screentone background with floral motifs, sparkle accents. Speech bubble with small lettering: "I didn't know you'd read them." Thought bubble at corner: "...I should have known." Clean ink lines, delicate shading, signature shojo eyes with multiple catchlights.
Tip: the two-bubble (speech + thought) composition stays clean.
34. 6-panel comic strip.
A 6-panel newspaper-style comic strip, landscape orientation, 3 panels over 3 panels. Recurring character: a rumpled cat in a lab coat. Panel 1: cat at a whiteboard labeled "DAY 47". Panel 2: cat stirring a beaker. Panel 3: beaker glowing. Panel 4: cat triumphant, arms raised. Panel 5: cat stares as the beaker fades to plain water. Panel 6: cat resigned, sips from the beaker, caption: "Coffee works too." Same cat in every panel, consistent shape and posture. Simple line art, light wash of color.
Tip: "same cat in every panel" is what keeps the consistency — say it explicitly.
35. Isekai-style portrait spread.
Isekai fantasy manga double-page spread, landscape. Left page: a young hero in mismatched armor standing at a cliff edge, wind blowing his hair, staring at an enormous floating dragon in the distance. Right page: the dragon's head in extreme close-up, one enormous eye filling the frame, scales lit by sunset. Ornate panel borders. Japanese onomatopoeia in vertical katakana "ゴオオオ" (deep roar) layered across both pages. Full-color illustration, painterly light.
Tip: correct vertical katakana for onomatopoeia is a 2026 GPT Image 2 capability.
36. Slice-of-life single panel.
Single-panel slice-of-life manga scene, wide aspect. A small ramen shop interior at night, three customers at the counter, rain streaking the window. The chef visible behind the counter, steam rising from a bowl. Warm orange light versus cool blue exterior. Speech bubble from the chef: "One more?" Speech bubble from a salaryman customer: "...yeah." Subtle screentone, hand-drawn feel, nostalgic atmosphere.
Tip: ramen shop with bilingual signage is a fun extension — try adding Japanese menu boards in the background.
37. Superhero action panel.
Superhero comic panel, vertical. A costumed hero in a blue-and-gold suit punching through a concrete wall, debris flying in a halo. Background: collapsing warehouse interior. Sound effect "THWACK!" in bold yellow-and-black onomatopoeia across the upper left. Caption box: "He didn't knock." Energy-line effects radiating from the impact. Inked lines, flat colors, classic Western superhero style.
Tip: Western-comic onomatopoeia follows different typography rules than manga — GPT Image 2 distinguishes.
38. Manga villain reveal.
Dramatic villain-reveal manga panel, vertical. A masked figure in a black cloak stepping out of shadow, half the face illuminated by a single overhead light. The visible eye is silver, pupil slit. Background: stone temple pillars, dust particles in the light beam. Speech bubble at the top, long-tailed: "You were always too predictable." Small screentone crosshatching on the cloak. Black ink on cream paper.
Tip: "half the face illuminated" is respected literally — the model does the lighting split.
39. Cooking manga how-to panel.
Cooking-manga instructional strip, 4 vertical panels. Panel 1: a chef character holding up an egg, caption box "Step 1: Crack gently". Panel 2: cracking the egg into a bowl, caption "Step 2: Let it fall". Panel 3: whisking, caption "Step 3: 30 seconds of wrist work". Panel 4: a finished tamagoyaki on a plate, caption "Step 4: Perfect every time?". Speech bubble in panel 4: "Maybe next time." Clean ink lines, light screentone, warm yellow accent on the egg and final dish.
Tip: numbered captions + single accent color across 4 panels — a durable template.
40. Manga cover illustration.
Manga volume cover illustration, A5 portrait. Title at top in stylized Japanese and English: "月の向こう" with smaller English "BEYOND THE MOON". Volume indicator: "VOL. 3" in a circular badge. Central illustration: two characters back-to-back, one holding a glowing pendant. Ornate border around the cover. Publisher logo bottom left. Barcode placeholder bottom right. Full-color painterly illustration with clean type overlay.
Tip: bilingual title rendering (Japanese + English) is the right kind of ambitious for GPT Image 2.
5. Multilingual Signage (10 prompts)
GPT Image 2 correctly renders Japanese, Korean, Chinese, Hindi, and Bengali in our tests. For other scripts, write the expected text in Latin transliteration AND in the target script — the model uses both as reference.
41. Tokyo ramen shop.
Photorealistic Tokyo ramen shop storefront at night. Main sign across the top in bold kanji: "一龍軒" (Ichiryuken). Subtitle in smaller Japanese below: "本格手打ち麺" (authentic handmade noodles). English below in 14pt: "Est. 1978". Red lantern with kanji "らーめん" hanging beside the door. Menu board visible in the window with three prices listed in yen (¥850, ¥950, ¥1,100). Warm interior light spilling onto wet pavement. Drizzle of rain. Shallow depth of field.
Tip: ¥ symbol, kana, and kanji all in one shot — this was impossible pre-GPT Image 2.
42. Seoul café.
Bright Seoul coffee shop storefront, daylight. Main hangul sign reads "하루 한 잔" (One Cup a Day). Smaller English under: "HARU · SINCE 2022". Menu board inside the window in hangul with prices in won: "아메리카노 ₩4,500", "카페라테 ₩5,000", "바닐라 라테 ₩5,500". Potted plants outside, chalkboard with a handwritten hangul greeting: "오늘도 좋은 하루 되세요" (Have a good day today). Glass door, clean minimal facade, natural light.
Tip: hangul renders at this level of fidelity only in GPT Image 2.
43. Delhi street food stall.
Vibrant Delhi street food stall, daylight. Hand-painted sign in Devanagari: "रामू की चाट" (Ramu's Chaat). English translation below in smaller letters: "RAMU'S CHAAT · SINCE 1962". Price list in Devanagari and English on a side board: "पानी पुरी · Pani Puri · ₹60", "आलू टिक्की · Aloo Tikki · ₹80", "दही भल्ला · Dahi Bhalla · ₹100". Bright colors, marigold garlands, stainless-steel bowls, customers gathered. Golden hour light.
Tip: Devanagari is a GPT Image 2 demo prompt — check each conjunct character.
44. Chinese teahouse.
Traditional Chinese teahouse sign, vertical orientation. Main characters in calligraphic simplified Chinese: "静心茶舍" (Quiet-Heart Tea House). Subtitle in smaller characters: "百年老字号" (Century-old establishment). English translation below: "Jing Xin Tea House · Est. 1923". Wood-grain background, brass lanterns flanking the sign, one red tassel hanging from the corner. Natural dusk light, soft warm tones. Photographic realism.
Tip: vertical character stacking for Chinese calligraphy is a test of structural layout adherence.
45. Bangkok night market stall.
Bustling Bangkok night market food stall. Sign in Thai: "ส้มตำป้าหน่อย" (Auntie Noy's Som Tum). English below in 18pt: "AUNTIE NOY'S SOM TUM". Handwritten menu board with 4 items in Thai and English, prices in baht ("฿80", "฿100", "฿120", "฿150"). Strings of bare bulbs overhead, steam rising from a wok, a crowd of customers in flip-flops. Motion blur of a passing tuk-tuk in the background. Golden-orange evening light.
Tip: Thai script is currently harder than CJK for GPT Image 2 — verify carefully.
46. Istanbul bazaar entrance.
Istanbul Grand Bazaar entrance, ornate stone archway. Sign in Turkish above the arch: "KAPALIÇARŞI · 1461". Secondary text in Ottoman Turkish calligraphy: "قپالی چارشی". English translation below in serif: "Grand Bazaar · Since 1461". Tourists entering in the frame, warm afternoon light. Intricate mosaic tile work on the archway. Editorial travel photography style.
Tip: Turkish diacritics (ç, ş) are now rendered reliably.
47. Dhaka bookstore.
Independent Dhaka bookstore storefront. Painted sign in Bengali: "বইয়ের ঘর" (House of Books). English below: "HOUSE OF BOOKS · Dhaka · 1989". Window display with 8 book covers arranged in a grid, each with visible Bengali and English titles. A handwritten sign on the door: "আজ খোলা" (Open today). Late afternoon light, mango tree casting shadow across the facade.
Tip: Bengali is one of the newer multilingual capabilities — the ligatures render cleanly.
48. Mexico City taquería.
Street-side taquería in Mexico City. Hand-painted sign in Spanish: "TAQUERÍA DOÑA LUPITA" in bold red lettering on white. Subtitle: "Desde 1978". Menu board with 6 items and prices in pesos: "Al pastor $30", "Suadero $30", "Lengua $35", "Barbacoa $40", "Tacos dorados $45", "Quesadillas $50". Steam rising from the plancha, a customer holding a paper plate. Bright midday sun, colorful papel picado strung overhead.
Tip: Spanish diacritics and the ñ are reliable — and the peso symbol renders correctly.
49. Quebec bakery.
Quebec City bakery storefront, snowy winter afternoon. Main sign in French: "BOULANGERIE DE LA MONTAGNE". Handwritten chalkboard outside: "Pain au chocolat · $4.50", "Croissant amande · $5.00", "Baguette tradition · $3.75", "Tarte au sucre · $6.00". Small bilingual notice on the door: "Ouvert · Open". Frost on the window, warm yellow light inside, a line of customers visible through the glass.
Tip: bilingual English/French signage works at both the sign and menu level.
50. Tel Aviv coffee shop.
Tel Aviv coffee shop exterior at golden hour. Sign in Hebrew (right-to-left): "קפה דיזנגוף" (Dizengoff Café). English below in 18pt sans-serif: "DIZENGOFF CAFÉ". Menu board with 4 items in Hebrew and English with shekel prices: "אספרסו · Espresso · ₪12", "קפה הפוך · Café Hafuch · ₪16", "איספרסו קר · Iced Espresso · ₪18", "קרואסון · Croissant · ₪14". Palm trees in the background, mid-century modernist facade, soft warm light.
Tip: right-to-left script direction is handled natively — no need to flip the image afterward.
Where to go next
If you built something good with any of these prompts, Oakgen's image gallery is shared across users (opt-in) — tag your generation public and it may show up in the model page's inspiration feed.
If you haven't yet: upgrade to annual Ultimate or Creator to get 30 days of GPT Image 2 included free, or read the launch post for the full rollout story. If you are deciding between GPT Image 2 and Nano Banana Pro for photoreal work, we tested both on 20 prompts.
And if you run a creator channel where prompt libraries land — Oakgen pays 25% on affiliate referrals for 6 months. Launch week is the right week to recommend a new tool honestly.
