LATO.2 open-sourced: factorized 3D mesh generation with controllable vertex count
rms80 · x · 2026-11-22
Researchers from Huazhong University of Science and Technology, Meshy AI (work done during internships), TUM and collaborators released and open-sourced LATO.2 ("Factorized 3D Mesh Generation with Vertex and Topology Flow"), already at 292 GitHub stars under MIT license.
The method factorizes mesh generation into two flows:
- V-Flow: generates vertex positions under a controllable vertex count — you can set the target count before generation;
- T-Flow: predicts connectivity (topology) from the realized vertices.
For image input, a TRELLIS-style coarse voxel scaffold is built first, then LATO.2 generates vertices and their connections. It supports part-based generation and topology-adaptive mesh editing. Project page: lohhhha.github.io/LATO.2.
More from Multimodal
- AI art project ACT III opens at Austria's Francisco Carolinum museum — Merzmensch · 2026-11-22
- Magnific Adds Kling Image 4.0 and Ideogram 4.6, MiniMax H3 Image Incoming — charis_ai · 2026-11-22
- RTX 3090 FP16/BF16 unlock on Windows cuts ComfyUI image gen time by up to 34% — notaneimu · 2026-11-22
- AI-generated podcast-style video of a voice that 'knows what missing someone feels like' — UnfilteredTalkOffice · 2026-11-22
- CapCut launches ChatGPT plugin that generates and edits videos inside chat — LearnWithBishal · 2026-11-22
- Indie creator ships AI anime short 'Start Again' with Higgsfield, Seedance 2.5 and Suno — SkankStar · 2026-11-22