I tried an AI-generated editable Three.js game — playable browser demo and an honest trade-off

[gamemakerH5]

I’ve been using SEELE AI to generate several 2D and 3D games from natural-language descriptions, and I wanted to share one browser-playable Three.js result with other web game developers.

 

Playable demo:

https://static.seeles.ai/media/3js/974610f7-0c59-4c8b-a7be-2a0ff7bcf20e/index.html

 

Demo video:

https://static.seeles.ai/media/demo_show/3d-frdj.mp4

 

The project is called Verdant Blade. It includes keyboard movement and sprinting, normal attacks, a special attack, dodging, health and energy UI, and enemy waves. The example above was generated without manual modification afterward. The generated code can be edited and exported.

 

My strongest positive impression is the quality and completeness of the initial result. It feels closer to a small playable prototype than a static scene or visual mock-up.

 

The main drawback in my experience is generation time. It can feel a little slow when testing several ideas in sequence. I currently see it as a way to get a more complete starting point rather than an instant rapid-iteration tool.

 

I’d be interested in how experienced Three.js developers evaluate generated projects like this. What would you inspect first—rendering performance, scene structure, asset loading, input handling, or maintainability—before deciding whether to continue developing the generated code?

 

Disclosure: I’m an independent SEELE AI user. I’m not employed by, contracted by, invested in, or otherwise affiliated with the company, and I received no payment, free credits, or other incentive for sharing this. The linked project is a public product example, not a project I personally generated.