Meshy T2: Fast Native Mesh Generation Framework via Flow Matching

Jiale Xu · hf · 2026-08-03

Meshy T2 is a fast native 3D mesh generation framework based on flow matching, designed to overcome the slow inference and error accumulation issues of mainstream autoregressive methods.

Core Architecture: Introduces a vertex-set mesh VAE that encodes a mesh into a continuous latent token per vertex and decodes vertices, edge connectivity, and face winding order in a single pass, preserving high-precision geometry and artist topology without vertex quantization or welding.
Generation Process: Employs a coarse-to-fine cascade of two flow-matching models. An image-conditioned voxel flow first sketches a coarse occupancy scaffold, followed by a mesh flow that populates per-vertex latent tokens conditioned on the image, scaffold, and a vertex budget.
Advantages & Performance: Supports interactive generation speed, effective face-count control, and native multi-part assets. Experiments show it achieves state-of-the-art geometric fidelity, completing end-to-end image-to-mesh generation in a median of 6 seconds, over an order of magnitude faster than autoregressive baselines.

Original post →

Meshy T2: Fast Native Mesh Generation Framework via Flow Matching

More from Multimodal