OpenAI GPT-6 turns photos into 3D scenes

Everyone gets access within a week; testing it via JetBrains, the author says it answers without the usual wall of padding and seems to know how a scene will look before it even finishes writing the code.

Author: Michael Kokin ·

New GPT-6 just dropped from OpenAI — everyone gets access within the week. I've been testing it hard already (thanks, JetBrains), and honestly, I'm impressed: I've seen and touched every big-name model, I'm hard to surprise, but this one managed it.

What impressed me most

A dacha near Odesa, in 15 minutes

Here's an example it put together in 15 minutes from a simple prompt — no spec, just "go do it, here are the photo and video files."

It's the dacha near Odesa where I spent a lot of time as a kid. I gave the model about 10 photos and videos, and in a couple of turns it built a 3D map — one I can use both for nostalgia and as an actual CS 1.6 map. What really got me were the textures: it cut frames out of the video itself and mounted them in frames, and turned one into a shower curtain.

Which tells me that at some point, any old family photos could be reconstructed into actual scenes — 3D or not, doesn't matter: the model just looks at the photos or frames in space, figures out where everything goes on its own, and builds the map it needs.