A small, detailed city rendered with Three.js, from a fountain plaza and café tables to an elevated track and a clock tower, all built in code.

Per the creator, the scene was made with Codex using GPT-6 Astra right after the Three.js r187 release, and the whole city uses no imported 3D models. The clip glides over shaded streets, trees, rooftop terraces and benches.

Why it matters: when an AI coding agent can assemble a scene like this from primitives alone, prototyping game worlds and architectural mockups starts to look more like describing a place than modeling it.

Credit: @zavrenn https://x.com/zavrenn/status/2116564845860360225/video/1