Leonxlnx/athenaion

Media & Design

A procedurally generated ancient Greek 3D world in three.js, useful as a reference for code-only scene building and offline rendering.

296
51
JavaScript
MIT
7 days old

Athenaion is best treated as a large interactive 3D art piece and rendering demo, not a reusable ancient-Greece toolkit or a game. If you want to see how far procedural geometry, shaders and an offline accumulation renderer can go in plain three.js, it is worth opening; if you need an editable asset library or a game world, look elsewhere.

The problem

A developer or graphics hobbyist wants a complete ancient Greek scene — terrain, temple, statues, water and vegetation — without sourcing models, textures or an asset pipeline. Normally that means assembling an art team or a store of assets, then building a scene around them. Here the scene itself is the deliverable: clone the repo and run it, and a full acropolis with lagoon, sea and forest is generated in the browser. What changes is that composition, geometry and look development all happen in code, with the reference images and renders in the repo as the target.

What's different

The core mechanism is that every visible element is generated procedurally: meshes, shaders, trees, statues and waterfalls, with statues described as signed distance fields and then meshed. The repo also ships an offline renderer using jittered accumulation for anti-aliasing, soft shadows and motion blur, plus AgX tone mapping and colour grading, so the same scene can be turned into 4K stills and video frames project instructions. The build was done by 55 subagents over roughly two days and about 147 agent-hours, with each specialist module rendering, comparing against a reference, and correcting in a loop reference comparisons.

That makes the repository interesting as a record of a specific process, not just a scene. The prompt and the art-bible/module-contract documents are checked in, so you can follow how the world was decomposed into parallel work session prompt. The trade-off is that the result is purpose-built: it closely tracks nine AI-generated reference images reference set, which means composition and detail choices serve those shots rather than general reuse. Expect a set of hero views, not a world you can walk through freely — the README describes switching between nine shots and orbiting with the mouse.

Who should try it

Good fit: someone studying three.js at scale, procedural geometry, or offline WebGL rendering who wants a complete working example with the code, prompts and references side by side. It is also useful if you want to understand how a large multi-agent coding session was split into module contracts. Poor fit: a game developer looking for a world to populate, or an artist who wants to modify meshes in a DCC tool — the scene is code, and the heavy geometry is built at runtime from scripts rather than committed as assets.

Requirements and limitations

Running it needs npm, then a local static server script; the first load is slow because heavy geometry such as terrain meshes, statues and the giant head is not committed and is built at runtime. The README offers a documented way to bake that geometry once ahead of time project instructions. Rendering stills and video requires Chrome or Edge, and the hardware claim is an integrated laptop GPU (AMD Radeon 780M) used for 4K rendering; no benchmark numbers are provided, so treat that as a feasibility signal rather than a performance guarantee.

Ongoing upkeep is minimal for a viewer, but the repo is a snapshot of one generated scene rather than a maintained engine. The author reports large token usage and an API-equivalent cost in the low thousands of dollars, which matters if you want to reproduce or extend the generation process rather than just view the result. The human time is not a substitute for a reusable asset pipeline, and the MIT license covers code, not the music used in the showcase video.

HotRepo's take

HotRepo's only recorded signal is 296 total stars and #25 on the JavaScript new-repository board on 2026-10-25 — a single dated snapshot, not a daily growth measurement repository. There is not enough comparable history to judge whether interest persists. Judge it as a technical showcase; if you intend to reuse the scene, first check whether the runtime-generated geometry and the nine fixed shots match what you actually need.

Written by AI from the project's README, updated October 25, 2026. The project's own docs are the source of truth.