TL;DR of the paper, generated automatically.

OmniLatent builds part-split 3D models from one image, skipping separate segmentation.

• Shared latent: geometry, deformation, texture and part IDs learned together.
• Editing: point prompts segment, refine and subdivide parts.
• Completion: fills in partial parts to keep whole objects consistent.