LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents

In this paper, we introduce LATO, a novel topology-preserving latent representation that enables scalable, flow matching-based synthesis of explicit 3D meshes. LATO represents a mesh as a Vertex Displacement Field (VDF) anchored on surface, incorporating a sparse voxel Variational Autoencoder (VAE) to compress this explicit signal into a structured, topology-aware voxel latent. To decapsulate the mesh, the VAE decoder progressively subdivides and prunes latent voxels to instantiate precise vertex locations. In the end, a dedicated connection head queries the voxel latent to predict edge connectivity between vertex pairs directly, allowing mesh topology to be recovered without isosurface extraction or heuristic meshing. For generative modeling, LATO adopts a two-stage flow matching process, first synthesizing the structure voxels and subsequently refining the voxel-wise topology features. Compared to prior isosurface/triangle-based diffusion models and autoregressive generation approaches, LATO generates meshes with complex geometry, well-formed topology while being highly efficient in inference.

Creating high-quality 3D digital objects quickly and efficiently is a major challenge in artificial intelligence and computer graphics. Typically, AI models struggle to cleanly define the underlying wireframe—or "topology"—of these 3D shapes. This often results in messy, irregular models that are difficult for 3D artists, game developers, or engineers to use without extensive manual cleanup. In this work, we introduce LATO, a new AI method designed to generate 3D objects with clean, artist-friendly structures right out of the box. LATO solves the wireframe problem by teaching the AI to explicitly understand how the individual points of a 3D object should connect to form a surface. Compared to previous methods, LATO is not only much faster at generating 3D shapes, but it also consistently produces complex, highly detailed models with well topology that are immediately ready for use in video games, virtual reality, and digital design.