Researchers use large language models to let users control the invisible back sides of 3D objects through text commands, solving a major limitation in single-image 3D generation.

Know3D enables text-prompt control over the hidden surfaces of 3D objects that aren't visible in single images

The technique leverages the world knowledge stored in large language models to generate realistic back-side details

This addresses a critical blind spot in current single-image 3D generation technology where unseen surfaces are difficult to control