dsebastien/countries-cities-landmark-images-skills

This repository is a small toolkit for generating polished, photo-realistic AI images of countries, cities, and global landmarks. The original motivation, mentioned in the README, was a school fair where the author wanted a single striking picture per country instead of a mood-board of disconnected tourist photos. The same images would also suit blog covers, presentation slides, or travel-style spreads. There are two main pieces, called skills. The first, travel-country-poster, generates one blended cinematic photo per country or city at golden hour, with the country's famous landmarks fused into a single frame and the national flag billowing in the upper-left corner. The second, world-landmark-globe, produces a hero image of planet Earth seen from low orbit with 3D miniature landmarks standing on each continent at their real geographic locations. The examples folder shows several finished images for places like Paris, Tokyo, Italy, and various globe versions. Under the hood, both skills call a small Node.js script in scripts/nano-banana.mjs, which is a thin wrapper around Google's Gemini 3 Pro Image model (also known as Nano Banana Pro). The script supports two modes: generate for making a brand new image from a prompt, and edit for changing an existing image while keeping most of it intact. Command-line flags choose aspect ratio, size (up to 4K), output path, and model. The README warns that this is a paid Google API, costing cents per 2K image, and that the API returns JPEG bytes even if you ask for a PNG. You need Node.js 18 or newer and a Google AI Studio API key, set as the GOOGLE_BANANA_API_KEY environment variable. The skill folders are written in Claude Code's SKILL.md format and can be dropped into a user or project skills directory, after which Claude Code triggers them on phrases like "make me an image for Japan". They can also be read as plain prompt-engineering documentation: each one contains a fill-in-the-blanks template, a landmark library per country, style rules (always golden hour, spatial positioning words, a long negative tail to avoid text overlays), and iteration recipes. The project is MIT licensed and the example images carry Google's C2PA content credentials.