Video generation and feed-forward 3D reconstruction usually operate in separate representational spaces. According to the arXiv paper "Latent Bridge: From Video Latents to 3D Geometry via Feature Alignment," this separation forces existing pipelines to decode generated latents into RGB images and then re-encode them visually to recover geometry.
The authors propose Latent Bridge, a method that connects video latents to 3D geometry through feature alignment. The provided abstract describes the goal as avoiding the need for RGB decoding and visual re-encoding, though the abstract text is truncated in the source, so architectural details and evaluation results are not covered here.
Since only one source is available, there are no conflicting claims to compare. The paper is listed as a new arXiv announcement under identifier 2610.28521v1.