Two Komorebi towns, one shared prompt

The only game-building instruction given to both

Both received only the Japanese text below, with no additional game-building instructions or supplied artwork. Creating the pixel art was part of the request. An English translation is provided for readers; it was not a second prompt sent to the models.

This is an exhibition of the result returned after one request. The accompanying development notes also describe implementation, checks and fixes during production. “One prompt” does not mean that no revisions happened during generation. Descriptions are based on the supplied code and records. Model names are the labels used for this experiment, and these two examples do not establish an overall ranking of the models.

ドット絵のキャラクターが日本の小さな町の中で生活している様子を眺める放置型のゲームを作ってください。ドット絵も作ってください。で基本HTMLで動くようにしてください。それぞれのキャラがある程度自律的に動いて、町の中で生活している感じを出してください。  町の人の数は10人ぐらいで、えっと日本の普通のコンビニとかスーパーとかあったりとか、車が走ってたりとかする小さな町、田舎町でイメージして作ってください。なんかそれぞれのキャラクターが交流したり買い物したりとかしてくれると嬉しい。 キャラクターの外見とか年齢とかもそれぞれ違う感じにしてください。  

Argon: explore the town and trigger small events

Komorebi Town Life lets you drag and zoom around the town. Select a resident to inspect their activity, mood and shopping bag, or follow them with the camera. Buttons can trigger a supermarket sale, new convenience-store sweets and passing vehicles.

Its accompanying Markdown describes choosing destinations from preferences and time of day, routes through crosswalks and the bridge, conversations between nearby residents, and items acquired while shopping. Canvas draws the people and town, with changing daylight, lit windows, streetlights and drifting petals. Sounds are synthesized in the browser.

The game logic and interface are contained in one HTML file. It loads Tailwind CSS and Google Fonts from external services, so it is not fully offline.

Sol: watch daily routines and an ongoing town

Komorebi Town places the whole-town view beside a resident list. Select someone to see their age, job, activity, energy, hunger, social state and frequent conversation partners. Its follow control places a marker on the person or the building they are in; the camera continues to show the whole town.

The development record describes decisions based on time, work, hunger, energy and weather, pauses between conversations, movement along paths, and pedestrian–vehicle coordination at crosswalks. Fixed background patterns reduce flicker, while age and clothing help distinguish the residents.

It runs from five HTML, CSS and JavaScript files without external images, audio or libraries. Town progress is saved locally in the browser, and the photo button exports an image. Conversations use prepared lines; the game does not contact an external AI service during play.

Try watching one resident in each town

Choose a resident and watch where they go, what they buy and whom they meet. Advance time to compare the lighting, then look at how each interface communicates the town and its people. The aim is to explore which interpretation feels like a place you would enjoy watching.

Descriptions are based on the supplied code and development records. No quality scores, performance rankings or generation-time comparisons are offered. Screenshots show the rendered town only, rather than Argon’s complete interface.