LATO.2: Topology-Aware 3D Mesh Generation
Hang Long · hf · 2026-07-14
The authors propose LATO.2, a factorized flow matching framework tailored for topology-aware mesh generation.
Problem
Existing methods often jointly model vertex geometry and connectivity within the same latent space. This entangles continuous geometry with discrete topology, frequently leading to:
- Vertex drift
- Surface tearing
Method
LATO.2 decouples generation into two stages:
- Vertex flow: Generates vertices first
- Connectivity flow: Generates connectivity conditioned on the generated vertices
Both stages are anchored to a shared coarse voxel scaffold. Two corresponding VAEs are designed so vertices achieve sub-voxel precision while discrete connectivity is embedded into a continuous latent space.
Additional Advantages
This factorization unlocks two distinct capabilities:
- Block-wise generation: The scaffold can be partitioned into chunks and synthesized independently, enabling the generation of higher-resolution meshes.
- Topology-adaptive editing: Modifying first-stage vertices automatically updates the connectivity accordingly, without requiring re-optimization.
The authors state that experiments confirm its geometric fidelity and connectivity quality surpass current state-of-the-art methods.
More from Multimodal
- Pablo Stanley shares a full AI video workflow using ChatGPT, Gemini, Runway and CapCut — jdjohnson · 2026-07-21
- Meta AI text input now lets users interleave images with text — ezyang · 2026-07-21
- ShotPlan adds learnable planning tokens for cinematic multi-shot video generation — Tele-AI · 2026-07-21
- Same prompt, Seedance 2 and Grok are compared on cinematic transformation output — LudovicCreator · 2026-07-21
- CG Chefs Showcases Retro Anime Style AI Video Generation — nicolascraske · 2026-07-21
- Night-party video demo uses Seedance 2.0, timecode prompts and 4K upscaling — gen_ericai · 2026-07-21