Developer demonstrates complex 3D camera model built with GPT-6 Astra

X user elvis has posted a series of demonstrations in which the language model GPT-6 Astra generated Three.js code for a 3D camera model comprising 122 component groups and 1,877 model parts, a process that took roughly three hours.

The project shows the model writing Three.js at a level sufficient for complex mechanical assemblies — work that typically demands expertise in both computer graphics and engineering. The developer said he had not realized Three.js could produce models of this sophistication.

In parallel, elvis released an interactive prototype guide for a "Button Box" also developed with GPT-6 Astra. The guide lets users rotate the enclosure, inspect component placement, zoom into individual parts, trace wiring, and locate purchase links for each component.

The developer plans to publish a detailed methodology within the coming week, alongside additional examples already shared. The write-up is expected to reveal the prompting strategy and the breakdown of the modeling task into a sequence of steps the model executes.

The demonstrations add to a growing trend of using large language models for graphics and engineering code, though they remain a single user's account without official benchmarks or comparisons to dedicated CAD tools. No technical details on GPT-6 Astra's architecture or its limits on extended modeling tasks have been disclosed.