GPT-6 Astra turns 9 crappy photos into a fully interactive 3D model of a room
YuXiang_IRVL · x · 2026-09-09
User rpnickson demonstrated feeding GPT-6 Astra just 9 casual photos of his studio: the model understood the spatial arrangement, stitched the shots together, and produced a fully interactive 3D model of the space. Reposter YuXiangIRVL calls it another mind-blowing 3D reconstruction example, with a link for others to try.
More from Multimodal
- Teknium: GPT-Image-2.5 now live in Hermes Agent via Codex and fal — alexcovo_eth · 2026-09-09
- ComfyUI users struggle with node conflicts as new image and video models ship daily — StacksGrinder · 2026-09-09
- Hyper3D WorldGen turns a single photo into a fully editable 3D scene — dr_cintas · 2026-09-09
- MovieGrid uses multi-grid post-training to make long-form multi-shot video generation coherent — Jiawei Mao · 2026-09-09
- Fudan Proposes Four-Level Framework for Agentic Visual Generation — Fudan-University · 2026-09-09
- VidaForge Open-Sources Infrastructure for Video Pretraining Data Recipes — VidaForge · 2026-09-09