Landmark Miniature · City Name in Clouds · Open Skill 公开技能
Landmark Miniature · City Name in Clouds
Landmark Miniature · City Name in Clouds
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/pm2y3u69.jpg
# Landmark Miniature · City Name in Clouds
Turn "the most recognizable landmark area of a city" into a portrait promo image: the landmark looks like an incredibly detailed physical tabletop model, shot with real side light and a tilt-shift-like shallow depth of field, while the city name floats above it as readable white vapor lettering. It produces **a single image** — there is no multi-stage flow.
## When to use
- The user gives a city name and wants a landmark miniature poster / city promo image.
- The user uploads a photo of a landmark area and wants a miniature model made from that photo.
- The user only wants to change one part: a different city, a different landmark area, a different reference photo, landscape/square instead of portrait, or different lighting or lens feel.
## Steps
1. **Collect the inputs**
- **City name (required)**: the vapor lettering has to be readable, so the city name must come from the user or be unmistakably legible on the reference photo; **never guess the spelling of a city name** — if unsure, ask only that one question ("Which city?").
- **Landmark / specific area (optional)**: if the user names a district (e.g. the Bund, Asakusa, Marina Bay), put it in the prompt; if not, write "the city's most recognizable landmark area" — do not invent a building that does not exist.
- **Reference photo (optional)**: a real photograph of the landmark area, used to keep the building massing and spatial relationships accurate.
- **National flag (optional)**: add it only when the city maps unambiguously to one country (Tokyo → Japan); if it is disputed or uncertain, **leave the flag out** — better to omit it than to fly the wrong one.
2. **Write the English prompt from the fixed template** (replace the `{}` placeholders; the template itself is this skill's style contract — do not change the style, do not add or drop elements)
```text
City promotional key visual of {CITY}: the most recognizable landmark area of {CITY} — {LANDMARK_OR_AREA} — rendered as an incredibly detailed physical miniature model (handcrafted scale-model kit: resin and plaster buildings, tiny painted roads, miniature trees and street furniture, believable terrain). The city name "{CITY_IN_LETTERING}" floats in the sky above the landmark as readable white-vapor typography, letters solid and clearly legible, edges softly dissolving into cloud. A subtle {COUNTRY} national flag floats just above the lettering. Strong natural side sunlight, warm and directional, raking across the model and casting long soft shadows; shallow tilt-shift-style focus — the landmark and the tangible terrain are tack sharp while the upper sky and the extreme foreground fall off softly, like a close-up macro photograph of a tabletop model on a real desk. Portrait 3:4 vertical composition: the landmark occupies the middle-lower part of the frame, the vapor city name sits in the upper part above it with generous breathing room, no other text anywhere in the image. Follow the reference photograph for the landmark's architecture, massing and spatial arrangement. Do not show a full city map, a satellite top view, a collage of cities, borders, frames, captions, logos or watermarks.
```
Conventions:
- `{CITY}` uses whatever form the user gave (write the city name in the user's language; for a Latin name use uppercase, e.g. `TOKYO`); `{CITY_IN_LETTERING}` matches the user's spelling of the city name — **do not mix languages**, do not split the letters, do not add decorative glyphs.
- `{LANDMARK_OR_AREA}` is filled in only if the user named it; otherwise write `its single most iconic landmark area`.
- When there is a reference photo, keep the last sentence `Follow the reference photograph ...`; when there is none, delete that sentence and keep the rest unchanged.
3. **Generate the image** (use `generate_image` only)
- **No reference image (text-to-image)**: `generate_image(prompt=<the template above>, model='t2i.img2.5', orientation=