Meshy T2: Fast Native Mesh Generation with Flow Matching
Meshy T2 reports median end-to-end image-to-mesh generation in 6 seconds.
The paper frames the speed gain against autoregressive mesh decoders, which it says are slow and prone to accumulated errors. Its system uses a vertex-set mesh VAE and a two-stage flow-matching cascade: first a coarse voxel scaffold, then per-vertex latent tokens. The authors say the design preserves geometry and artist-style topology without vertex quantization or welding, while allowing face-count control through a requested vertex budget. Code and weights are listed as forthcoming. HF Daily Papers' note
The paper frames the speed gain against autoregressive mesh decoders, which it says are slow and prone to accumulated errors. Its system uses a vertex-set mesh VAE and a two-stage flow-matching cascade: first a coarse voxel scaffold, then per-vertex latent tokens. The authors say the design preserves geometry and artist-style topology without vertex quantization or welding, while allowing face-count control through a requested vertex budget. Code and weights are listed as forthcoming. HF Daily Papers' note
score 6