AnalysisAI ModelsNovember 7, 2026

OmniLatent encodes 3D parts in a shared latent for image-to-asset editing

Read original source →arxiv.org

OmniLatent jointly encodes geometry, deformation, appearance and part IDs in one latent space, generating part-aware 3D assets from a single image. It supports segmentation, refinement, subdivision and amodal part completion via point-prompt adapters.

1 source

More stories today

Open the live feed