GAE: Geometry-Native Latent Space for 3D-Consistent Generation
September 20, 2026
GAE reparameterizes geometry foundation model features into a compact latent space that is jointly decodable to appearance, depth, cameras, and point maps. This approach ensures 3D consistency in visual generation by encoding cross-view structure directly into the latent representation.
HOW THIS AFFECTS YOU
●
builderYou can build more stable 3D-consistent video and image generators using this representation.
●
researcherThis shifts the focus from appearance-centric latents to geometry-native representations for modeling.