3D Mesh Generation

3D mesh generation studies how to create or edit surfaces made of connected vertices and faces while controlling both shape and topology. Autoregressive models serialize vertices and faces into tokens, while diffusion and flow-matching methods generate meshes in parallel or through continuous latent representations; topology-aware fields and coarse-to-fine schemes offer other ways to represent connectivity. Models can be conditioned on text, images, point clouds, or motion, with evaluations spanning object-mesh datasets such as Objaverse and Toys4K as well as specialized settings like garments, anatomy, and engineering. The central challenge is balancing geometric fidelity, valid and useful topology, controllability, and efficient generation, especially as meshes grow more detailed or require artist-ready structure.