Automated Construction and Texturing

This first-party copy preserves Eric’s original AdvisoryHour/Substack post for the research section of EricRhea.com. The original remains available at advisoryhour.substack.com.

Generated retail-store scene used as an automated 3D construction and texturing test. — Automated 3D factory: the generated retail environment used to test construction and texture workflows.

Automated 3D factory

I mentioned in prior days how building Wrigley Field showed me a new path to building 3D geometry. In this experiment, I took that idea and focused on texturing automation using ChatGPT-5.5. Could the model build a clothing retail store? How would I solve texturing in a performant way? What are the “big ideas” in using AI to build out geometry?

The beautiful insight with a retail clothing store is that it allows you to quickly identify all the problems in your texturing process. It could be aesthetic quality, memory budgets, scale, or other dimensional concerns. A clothing retail store allows you to see every mistake, fast.

Generated clothing retail store overview showing repeated garments, racks, shelves, and storefront materials. — The retail scene makes texture and material mistakes visible quickly: garments, shelves, floor, lighting, and scale all compete in the same frame.

How the world was constructed

In this case, the world wasn’t constructed using the standard pairing of say Codex or Claude with ThreeJS. Instead, this world is constructed with a Blender pipeline using the Blender MCP. I’m consistently surprised at both the speed of this and how well it works overall. I’m getting effectively low-poly 3D models for free on any topic. This is incredibly liberating—I can drop into Blender still, and those shortcut keys are forever burned into my mind.

The thing is that it’s very difficult to pull off this scene—it’d take some care and some work as a human to build both the 3D geometry as well as the texture components.

Close view of generated store materials showing floor tiles and local-AI texture detail. — Small material decisions—floor tiles, texture scale, and object-level variation—are where the pipeline either starts to feel grounded or starts to collapse.

There’s a lot of subtle details that are hard to see unless you zoom into a given component. For example, the floor tiles are simple and nice. The really neat thing? These textured images are generated using local AI. I’m very happy with the result.

Generated shirts in the retail scene showing varied textile textures and garment surfaces. — Repeated shirts become a fast test for whether texture variation can stay readable without hand-authoring every garment.

In order to test the texturing of different clothing, I had 5.5 build shirts in the scene. I wasn’t sure if the cloth modifier would be used or what to expect. Perhaps with a little elbow grease I can get a better cut of shirt, but that wasn’t the purpose of this test. Instead, consider the texture variety here. Each shirt is a subtle, unique texture variation.

Texture atlas view used to create varied retail clothing materials in one image. — The key idea: texture atlas discipline, combined with normal maps, creates variety without exploding the asset budget.

The key idea is the texture atlas. This allows for a significant variety of textures when combined with a normal map and it all fits in one image.

Navigable 3D retail environment generated for browser exploration. — The scene is intended as a navigable environment, not just a single hero render.

The entire scene is navigable

The entire scene is navigable. Feel free to fly around and check it out. The link is in my research subfolder over at my website. Next I’ll be scaling this out even further. How big of a scene can I make using these new techniques?

Open the Flythrough Viewer Open Retail Materials Test

Gallery of generated texture samples for use in 3D projects. — A texture-gallery frame from the original post, preserved locally for project reference.

Texture gallery

Here’s a gallery of textures for you to use in your projects.

Thanks for reading. Subscribe for free to receive new posts and support my work.