GPT-6 × img2threejs v2.2
GPT-6 launched while I was halfway through building img2threejs 2.2, so I ran an early comparison between Luna, Sol, and Astra.
Same reference image.
Same prompt.
Same unfinished pipeline.
Skill:
“Use the img2-environment skill to recreate the room from this reference image: [link]”
v2.2 is currently around 50% complete, but the differences in how each model interprets the same scene are already clear. Sol’s visual understanding stood out to me in this test.
The goal for v2.2 is to push reference-to-scene fidelity as close to 99% as possible.