world-landmark-globe

name	world-landmark-globe
description	Generates a photorealistic editorial "hero" image of planet Earth viewed from low orbit, with iconic 3D miniature landmarks rising from each continent at their real geographic locations. Cinematic golden-hour lighting on the visible hemisphere, soft starlit space margin, optional aurora borealis and polar elements. Magazine-cover quality.
when_to_use	Use when the user asks for a "world image", "globe with landmarks", "Earth with monuments", "world map cover", "all countries in one image", or any single image that should represent the whole world / multiple continents at once (cover art, school-fair centerpiece, presentation hero slide, travel banner).
model	claude-haiku-4-5-20251001
effort	low
argument-hint	["default","--zoom close","--landmark-density high"]
allowed-tools	Read Bash Glob Grep AskUserQuestion
metadata	{"kind":"generator","capability":"media.image.world-globe","effects":"external","tier":"composed","created":"2026-05-19T00:00:00.000Z","bundles":["scripts/nano-banana.mjs"],"related":["travel-country-poster"]}

World Landmark Globe Generator

Produces one large hero image of Earth with 3D miniature landmarks placed at their actual geographic locations — like a Monopoly board wrapped onto a globe. Sister skill to [[travel-country-poster]]; share the same golden-hour palette so a country poster and the globe can sit next to each other in the same display without clashing.

Pipeline

Same script as the country-poster skill — the bundled scripts/nano-banana.mjs — with these settings:

• Model: gemini-3-pro-image-preview (Nano Banana Pro). Required for the architectural detail and proper spherical perspective. Flash collapses landmarks into mushy blobs at this scale.
• Aspect: 16:9. The globe centers in the frame; aspect leaves room for stars and the curved limb.
• Size: 2K (2752×1536). The detail-per-landmark only pays off at 2K; 1K loses the architectural recognizability.
• Output: JPEG (auto-renamed by the script).

Critical Style Rules

1. Globe fills 90-95% of the frame. Not 60% in a sea of black. Phrase: "the Earth fills approximately 90 to 95 percent of the frame with only a thin sliver of starry space visible around the curved edge".
2. Landmarks are HERO-SIZED, not pinpoints. Phrase: "3D miniature landmarks rise dramatically from each continent like detailed museum-model architecture, each placed at its real geographic location on the globe surface, scaled large enough to be clearly recognizable from this viewpoint". If the result looks too small, regenerate with "each 3D miniature should appear roughly twice as large as in the reference, dominating its country, clearly recognizable at first glance".
3. Geographic placement is named explicitly. For each landmark write "[NAME] in [COUNTRY]" or "[NAME] on the [REGION] coast". The model uses this to anchor placement on the curved surface.
4. Continents grouped, listed in caps in the prompt. Use NORTH AMERICA:, EUROPE:, ASIA:, etc. as in-prompt section headers — the model treats these as scene-grouping cues and lays out the continent landmarks coherently.
5. Same cinematic golden hour as the country posters. Phrase: "Cinematic golden hour sunlight catches the visible hemisphere from the upper right, casting warm light on each landmark while the night side fades into soft starlit darkness on the far edge". This lighting choice is what makes globe + country posters visually consistent.
6. Atmospheric realism around the limb. Phrase: "soft atmospheric blue glow on the curved horizon". Without this, the globe looks flat and CGI-fake.
7. Polar bonuses (optional but recommended). Aurora borealis over the Arctic, penguin colony / iceberg on Antarctica. They fill what would otherwise be empty white poles.
8. No flags. The flag rule from the country-poster skill is REVERSED here — explicitly say "no flags" in the negative tail. Flags clutter a continent-density composition.
9. Negative tail (extended). "No cartoon, no flat illustration, no text overlays, no captions, no labels, no logos, no watermarks, no flags". The "no labels" is critical — without it, the model writes "HOLLYWOOD" or "EIFFEL" next to landmarks.
10. Iterate by edit, not regenerate. Adding landmarks, zooming in, scaling up — always via edit against the previous version. Regenerating from scratch loses the continent layout you already liked.

Landmark Library (geographic placement)

Use these as a starter set. Hero version = 25-30 landmarks; light version = 15-18. Beyond ~35 the model starts mashing adjacent ones together.

Region	Landmark	Real-world location
North America	Statue of Liberty	NYC east coast
	Empire State Building	NYC
	Hollywood sign	Los Angeles, US west coast
	Golden Gate Bridge	San Francisco
	Mount Rushmore	South Dakota, central US
	Niagara Falls	US/Canada border
	CN Tower	Toronto
	Chichen Itza pyramid	Yucatán, Mexico
South America	Christ the Redeemer	Rio de Janeiro, Brazil
	Machu Picchu	Peru
	Iguazu Falls	Brazil/Argentina border
	Easter Island moai	small island in the Pacific off Chile
Europe	Eiffel Tower	Paris, France
	Atomium	Brussels, Belgium
	Leaning Tower of Pisa	Pisa, Italy (call out "visibly tilted")
	Colosseum	Rome, Italy
	Big Ben + Tower Bridge	London, UK
	Sagrada Família	Barcelona, Spain
	Brandenburg Gate	Berlin, Germany
	Acropolis / Parthenon	Athens, Greece
	Saint Basil's Cathedral	Moscow, Russia ("colourful onion domes")
	Dutch windmills + tulip fields	Netherlands
	Norwegian fjords	Norway (optional)
Africa	Pyramids of Giza + Sphinx	Egypt
	Mount Kilimanjaro	Tanzania
	Table Mountain	South Africa
	Victoria Falls	Zambia/Zimbabwe border ("with rainbow mist")
	Maasai village (optional)	Kenya
Asia	Taj Mahal	India
	Great Wall of China	northern China ("long winding section")
	Mount Fuji + Tokyo Tower	Japan
	Burj Khalifa	Dubai, UAE ("towering")
	Petra	Jordan ("carved rock facade")
	Angkor Wat	Cambodia
	Petronas Twin Towers	Malaysia
	Hagia Sophia	Istanbul, Turkey ("with minarets")
	Forbidden City pagodas	Beijing, China
Oceania	Sydney Opera House	Sydney, Australia
	Uluru	central Australia ("red sandstone")
	Mount Cook	New Zealand ("snow-capped")
	Great Barrier Reef	off northeast Australia ("turquoise patch with coral atolls")
	Polynesian outrigger	Pacific near Tahiti
Polar / atmospheric	Aurora borealis	over the Arctic, top of globe
	Antarctic ice shelf + penguin colony	bottom of globe
	Milky Way band	background space, faint

Prompt Template — Initial Generation

Photorealistic editorial illustration of planet Earth viewed from low orbit, the globe rendered with realistic landmasses, deep blue oceans, swirling white cloud cover, and a soft atmospheric blue glow on the curved horizon. The Earth fills approximately 90 to 95 percent of the frame with only a thin sliver of starry space visible around the curved edge. Iconic 3D miniature landmarks rise dramatically from each continent like detailed museum-model architecture, each placed at its real geographic location on the globe surface, scaled large enough to be clearly recognizable from this viewpoint. Cinematic golden hour sunlight catches the visible hemisphere from the upper right, casting warm light on each landmark while the night side fades into soft starlit darkness on the far edge.

NORTH AMERICA: [list 4-6 landmarks "[NAME] in [LOCATION]"]
SOUTH AMERICA: [list 2-4]
EUROPE: [list 6-10 — this is the densest continent, give it weight]
AFRICA: [list 3-5]
ASIA: [list 6-9]
OCEANIA: [list 3-5]

POLAR / ATMOSPHERIC: aurora borealis curling over the Arctic near the top of the globe, a small Antarctic ice shelf with a penguin colony near the bottom.

All landmarks are photorealistic 3D models with sharp architectural detail, properly oriented on the curved spherical surface, casting soft shadows on the terrain beneath them. Subtle atmospheric haze blends closer landmarks with more distant ones. Background is deep cosmic space with scattered stars and a faint Milky Way band. The composition is balanced, with the globe filling most of the frame and the curvature clearly visible. Ultra detailed, razor-sharp focus, professional editorial photography meets photoreal 3D rendering, color graded with rich cinematic tones, 4K, magazine cover quality. No cartoon, no flat illustration, no text overlays, no captions, no labels, no logos, no watermarks, no flags.

Run Command

The repo ships the generator at scripts/nano-banana.mjs. From the repo root:

node scripts/nano-banana.mjs generate "<prompt>" \
  --aspect 16:9 --size 2K \
  --model gemini-3-pro-image-preview \
  --output "<dir>/world_globe.jpg"

Iteration Recipes (tested in this directory)

Use edit against the previous version. Single focused instruction beats sweeping rewrites.

• Zoom in / pull the camera closer:

  "Re-render with the camera zoomed in closer so the Earth fills approximately 90 to 95 percent of the frame with only a thin sliver of starry space around the curved edge. Keep the same lighting and landmarks in the same positions."

• Make landmarks bigger (the most common ask):

  "Re-render keeping the same zoom, lighting, and landmark positions. SCALE UP every landmark substantially — each 3D miniature should appear roughly twice as large, dominating its country, clearly recognizable at first glance, with crisp architectural detail. Treat each landmark like a hero element rather than a small decoration."

• Add missing landmarks (group them by continent in caps, just like the original template):

  "Keep all existing landmarks in the same positions. Add the following landmarks at their real geographic locations: EUROPE: the Eiffel Tower in Paris, the Atomium in Brussels, the Leaning Tower of Pisa in Italy (visibly tilted)... ASIA: the Petronas Twin Towers in Malaysia..."

• Continent missing a key icon (e.g., Europe looks underweight):

  "Europe needs more density. Add prominent versions of the Eiffel Tower, Atomium, Leaning Tower of Pisa, Colosseum, Big Ben, and Sagrada Familia, each clearly recognizable. Keep the rest of the globe unchanged."

• Reduce or remove polar effects:

  "Remove the aurora borealis and penguin colony. Show clean polar ice instead."

• Different lighting (e.g., night side dominant):

  "Rotate the lighting so the night side faces the camera with city lights visible across the continents, golden glow on the right limb only. Keep all landmarks in place."

Workflow

1. Decide density. Hero = 25-30 landmarks. Light = 15-18. Ask the user if unsure.
2. Decide rotation. Default leaves Europe/Africa centered (Africa-front view). Mention this — they may want the Americas or Asia-Pacific facing camera. (If different, swap continents to the foreground in the prompt and reorder the continent blocks.)
3. Fill the template. Group landmarks by continent. List Europe first if Europe-front; list North America first if Americas-front.
4. Single generate call. Don't try to add landmarks in the same generation as zooming — the model gets overwhelmed and drops elements. Two passes (generate → edit-to-add-more) is more reliable.
5. Always Read the output file so the user sees it.
6. Iterate via edit. Use the recipes above.

Don'ts

• Don't use Flash. The architectural detail in the miniatures requires Pro.
• Don't omit the geographic placement strings (in Paris, in Tanzania). The model can't infer them from the landmark name alone — it'll dump everything in random spots.
• Don't request flags. They clutter the composition; the country flags belong on the country-poster images, not the globe.
• Don't try to fit > 35 landmarks. They collapse into each other.
• Don't omit "no labels" in the negative tail. The model will write "HOLLYWOOD" or "EIFFEL" otherwise — these are mostly suppressed by the tail but creep back in if you forget it.
• Don't expect symmetric continents — Europe has way more landmarks than Australia. Embrace that asymmetry in the prompt rather than padding minor continents.

Reference Outputs

Examples produced in this directory with this style:

• world_globe.jpg — first pass, default zoom (~80% globe-fill, ~22 landmarks)
• world_globe_v2.jpg — zoomed in, denser (~30 landmarks)
• world_globe_v3.jpg — biggest landmarks, Europe-emphasis pass; the canonical hero version

Any of these can be passed as the edit reference for further iterations.

Prerequisites

• GOOGLE_BANANA_API_KEY set in env.
• Node.js 18+ (the bundled scripts/nano-banana.mjs uses native fetch).