← sources

From One Image to an Explorable 3D World with GPT-6 Astra, Three.js, and Hyper3D

source: reddit (reddit:ChatGPT)published 2026-10-03 00:47ingested 2026-10-04 02:35

retained text

Started with a single image of a small winter scene and rebuilt it as an interactive 3D environment, including the cabin, trees, rocks, character, and surrounding landscape.

img2threejs analyzes the image and breaks it down into individual elements. Hyper3D provides the 3D references, while GPT-6 works through the img2threejs pipeline to reconstruct and assemble everything in Three.js, handling placement, proportions, lighting, and interactive behavior.

You can walk around, leave footprints in the snow, open the gate, adjust the time of day, and place a campfire beside the cabin after dark. The original image becomes a small world you can actually explore!

Hyper3D deserves a shoutout here. Having solid 3D references makes it much easier to translate the image into a cohesive scene, especially when you’re working with a whole environment full of different objects. It’s a useful bridge between the initial concept and a world you can actually explore.

Still an early showcase, so the reconstruction isn’t perfect and this isn’t a finished game. But it shows how much you can build from a single image.

What would you compare first: the object shapes, scene layout, or lighting?

document fetched 2026-10-04 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1wwa03c/from_one_image_to_an_explorable_3d_world_with