SpatialCrafter enables consistent video gen from single images via 3D proxies

kwangmoo_yi · x · 2026-08-28

SpatialCrafter introduces a framework using generative 3D proxies to build explorable scenes from single images. It combines 3D scene models with video models to refine renderings, addressing hallucinations and drift for high-fidelity, controllable generation.

Original post →

More from Multimodal

Multimodal channel →