Simon Willison has launched a new browser tool that renders 3D files directly in a web page. He used this utility to display a complex model of a Fabergé egg themed around the Pluribus TV show. The project began when he asked ChatGPT Images 2.5 to generate a concept photo of the egg. He then fed that image into a local Blender instance running GPT-6 Astra to create the actual 3D geometry. The AI coding agent spent nearly 18 minutes processing the request before outputting several .blend files. Willison integrated the resulting model into his existing viewer, allowing anyone to inspect the file without downloading it first.
This workflow demonstrates a practical method for combining generative image models with local 3D rendering engines. It shows how an LLM can act as a bridge between a visual prompt and a complex technical file format. The process also highlights the computational cost involved when running large models locally on a personal machine.
* The generation process took 17 minutes and 51 seconds.
* The final model contains over 1.4 million triangles.
* The viewer runs entirely in the browser via jsDelivr.
