New York, Nara and Chongqing, every building with its rooms, generated in your browser from a script of a hundred-odd kilobytes.
The first load takes a couple of minutes: every brick, roof tile, tree, room and note is made here, now. Nothing is downloaded but the engine and its script: 2 MB of WebAssembly, with a shader translator the macOS build doesn't need (the title card's count is that build's size). Later visits are quicker: this browser keeps the soundtrack and its compiled shaders.
To load faster next time, this browser keeps the soundtrack it composed (about 95 MB, in this site's storage) and its own copies of the compiled shaders.
The shaders go with the browser's cached files (Settings, Privacy, Clear browsing data, Cached images and files). It also registers a small service worker, which stores nothing: it adds the two headers its threads need (removed with the site's data, in the page's site settings).
Or press Tab to fly around the last city.