How Guojiaditu was made

Guojiaditu began as an experiment in presenting a country's geography, history, food, landmarks, festivals, clothing, and contemporary life as one approachable field guide. The aim was not to build another encyclopedia. It was to explore whether carefully structured writing and illustration could help someone form a memorable first impression of a place.

Building the first atlas

The original collection was assembled with a generation pipeline powered by Google Gemini. For each location, a structured editorial brief asked for a short overview and tightly constrained sections covering geography, a pivotal historical episode, landmarks, cuisine, festivals, traditional dress, and current themes. Separate visual briefs produced the illustrated scenes that accompany each entry.

The generated material was stored, reviewed through the website, and reused on later visits. Early versions of Guojiaditu could also create entries for regions and cities when a visitor requested them. That made the project useful for experimentation, but it also made the public website dependent on accounts, databases, background workers, and live model calls.

Why the site changed

Once the country collection had been assembled, live generation was no longer necessary for the experience we wanted to preserve. Guojiaditu therefore became a country-level atlas: a fixed, versioned collection that anyone can browse without signing in. Country pages are generated ahead of time so their text, links, and metadata are available directly to people, search engines, and other readers of the open web.

What remains human

Generative tools helped create the initial drafts and imagery, but the structure, constraints, product design, selection of subjects, and decision to publish a durable country collection were human choices. The resulting pages should be read as an illustrated introduction, not as a substitute for local voices, primary sources, or current reporting. As with any editorial atlas, details may need correction as names, borders, and public life change.

The current system

The public website no longer creates content on demand. It does not require a user account, and it does not send country requests to an AI model. Text and page metadata are stored as versioned static files; media is served from object storage. Lightweight, consent-aware analytics help us understand whether the collection is being found and used without bringing the old account and visitor database back into the product.