Mesh Generation¶
3D Generation
Open Source¶
-
ActionMesh: Animated 3D Mesh Generation with Temporal 3D Diffusion Discovered by AI
2026
A feed-forward model that generates animated 3D meshes from a single video, text description, or static 3D mesh. It extends 3D diffusion models with a temporal axis to produce topologically consistent, rigging-free 4D animations.
4danimationdiffusionmesh -
FACE: Face-based Autoregressive Mesh Representation
2026
A face-based autoregressive representation for high-fidelity and efficient mesh generation, predicting meshes face-by-face to improve detail and scalability.
autoregressivemeshrepresentation -
LATO Discovered by AI
2026
A topology-preserving latent representation and flow matching framework for explicit 3D mesh generation, representing meshes as a Vertex Displacement Field (VDF) and using a sparse voxel VAE to compress the signal into structured, topology-aware voxel latents.
flow-matchingmeshtopologyvae -
LATO.2 Discovered by AI
2026
A factorized flow matching framework that decomposes 3D mesh generation into vertex flow and topology flow, resolving the entanglement of continuous geometry and discrete connectivity.
flow-matchingmeshtopology -
Mesh BDF Discovered by AI
2026
Introduces the Barycentric Dominance Field (BDF), a continuous representation that enables native 3D mesh generation with high face counts, high vertex resolutions, and texture support, overcoming the limitations of autoregressive models.
barycentricgenerationmesh -
Mesh-Pro: Asynchronous Advantage-guided Ranking Preference Optimization for Artist-style Quadrilateral Mesh Generation Discovered by AI
2026
An end-to-end 3D mesh generation framework that introduces an asynchronous online reinforcement learning system tailored for 3D assets. It utilizes Advantage-guided Ranking Preference Optimization (ARPO) and a diagonal-aware tokenization scheme to produce high-quality, artist-style quadrilateral-dominant meshes.
arpomeshquad-meshreinforcement-learning -
MeshFlow (Equivariant Flow Matching) Discovered by AI
2026
A generative model that directly synthesizes triangle meshes as triangle soups using equivariant optimal-transport flow matching, respecting the permutation symmetries of vertices and faces without requiring serialization.
equivariantflow-matchingmesh -
MeshFlow (Facebook Research) Discovered by AI
2026
A method for generating artist-like 3D meshes using a Variational Autoencoder (MeshVAE) and a flow-based Diffusion Transformer, avoiding the quadratic scaling and quantization errors of autoregressive next-token prediction.
diffusionmeshtransformer -
2026
Artist-style mesh generation that assembles meshes from local patches ('mosaic') to scale to high face counts while preserving clean, usable topology.
artist-meshautoregressiveretopology -
Meshy T2: Fast Native Mesh Generation with Flow Matching Discovered by AI
2026
A fast native mesh generation framework built on flow matching that bypasses slow autoregressive serialization and iso-surface extraction, enabling high-quality, multi-part 3D mesh generation from a single image in seconds.
3d-generationflow-matchingmesh -
Modly Discovered by AI
2026
A local, open-source desktop application for Windows, Linux, and macOS that allows users to generate 3D meshes from images entirely on their local GPU using state-of-the-art open-source models.
desktop-appimage-to-3dmesh -
2026
An artist-mesh / re-topology generation method aimed at producing clean, animation-ready mesh topology from generated or scanned geometry (SIGGRAPH 2026).
artist-meshretopology -
Strips as Tokens: Artist Mesh Generation with Native UV Segmentation Discovered by AI
2026
An autoregressive mesh generation framework that uses a token ordering strategy inspired by triangle strips. By constructing the sequence as a connected chain of faces that explicitly encodes UV boundaries, it naturally preserves the organized edge flow and semantic layout characteristic of artist-created meshes, supporting both triangle and quad meshes with native UV segmentation.
autoregressivemesh-generationtopologyuv-unwrapping -
2025
Creates artist-style meshes with an autoregressive model trained with reinforcement learning to improve mesh quality and structure.
autoregressivemeshreinforcement-learning -
2025
A 3D shape VAE with a salience-aware point sampling strategy that preserves geometric detail, serving as a latent backbone for native 3D diffusion generation.
diffusionrepresentationsamplingvae -
DSO: Aligning 3D Generators with Simulation
2025
Aligns 3D generators with simulation feedback so generated objects are physically sound (for example, standing stable under gravity) rather than only visually plausible.
alignmentphysicsstability -
2025
A contouring / mesh-extraction method that faithfully recovers sharp surface detail from implicit fields, supporting high-fidelity 3D shape generation.
isosurfacemesh-extractionrepresentation -
2025
Efficient artist-style mesh generation that decouples geometry into components to shorten token sequences and speed up autoregressive mesh synthesis.
artist-meshautoregressivemesh -
2025
Accelerates VecSet-based 3D shape generation with a lightning vecset decoder and reconstruction optimizations, cutting mesh generation latency substantially.
accelerationfastvecset -
2025
Training-free generation of articulated 3D objects with movable parts and joints by repurposing a pretrained 3D diffusion model, requiring no articulation dataset.
articulateddiffusiontraining-free -
FreeMesh: Boosting Mesh Generation with Coordinates Merging Discovered by AI
2025
A plug-and-play tokenization technique called coordinate merging that optimizes mesh serialization sequences, significantly boosting the efficiency and quality of autoregressive mesh generation.
autoregressivemeshtokenization -
2025
Represents 3D geometry as a distribution learned with a diffusion/flow model over surface points, providing a robust neural shape representation for generation.
diffusionimplicitrepresentation -
2025
Performs amodal 3D part completion, generating complete part geometry including occluded regions for decomposed 3D objects.
amodaldiffusionpart-based -
Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material Discovered by AI
2025
Presents Hunyuan3D 2.1, a scalable 3D asset creation system that generates high-fidelity 3D meshes from images with production-ready Physically-Based Rendering (PBR) texture synthesis. It features a complete open-source framework including a VAE encoder and training code.
3d-generationdiffusionmeshpbr -
iFlame NEW Discovered by AI
2025
A Transformer-based architecture for 3D mesh generation that interleaves full and linear attention mechanisms to achieve sub-quadratic computational scaling while capturing global surface dependencies for high-resolution meshes.
3d-reconstructionlinear-attentionmesh-generationtransformer -
MAGE: Single Image to Material-Aware 3D via the Multi-View G-Buffer Estimation Model Discovered by AI
2025
A framework for material-aware single image-to-3D generation that leverages inspiration from traditional computer graphics deferred rendering pipelines. It introduces a multi-view G-buffer estimation model to simultaneously predict albedo, normal, depth, metallic, and roughness maps, producing high-quality 3D meshes with rich material properties.
g-bufferimage-to-3dmesh-generationpbr-textures -
Mesh-RFT: Enhancing Mesh Generation via Fine-grained Reinforcement Fine-Tuning Discovered by AI
2025
A fine-grained reinforcement fine-tuning framework for 3D mesh generation that employs Masked Direct Preference Optimization (M-DPO) to enable localized refinement via quality-aware face masking.
fine-tuningmeshreinforcement-learning -
MeshCraft: Exploring Efficient and Controllable Mesh Generation with Flow-based DiTs Discovered by AI
2025
A framework for efficient and controllable 3D mesh generation using flow-based diffusion transformers (DiTs), bypassing the slow token-by-token prediction of autoregressive methods.
diffusionditmesh -
MeshLLM Discovered by AI
2025
A framework that leverages large language models to understand and generate text-serialized 3D meshes, utilizing a Primitive-Mesh decomposition strategy to preserve structural information.
generationllmmesh -
MeshRipple: Structured Autoregressive Generation of Artist-Meshes Discovered by AI
2025
A structured autoregressive mesh generation framework that expands a mesh outward from an active generation frontier like a ripple on a surface, preserving long-range geometric dependencies and avoiding holes or fragmented components.
autoregressivegenerationmesh -
2025
Efficiently aligns 3D diffusion models using 2D rewards, improving generated geometry quality through reward-based post-training (NeurIPS 2025).
alignmentdiffusionreward -
Nautilus: Locality-aware Autoencoder for Scalable Mesh Generation Discovered by AI
2025
A locality-aware autoencoder framework designed for scalable 3D mesh generation, enabling direct modeling of continuous surface quality and producing high-quality, clean meshes.
autoencodermeshreconstruction -
PartCrafter Discovered by AI
2025
A structured 3D generative model that jointly synthesizes multiple semantically meaningful and geometrically distinct 3D meshes from a single RGB image without relying on pre-segmented inputs.
diffusionimage-to-3dmesh -
2025
Learns a feedforward part feature field on 3D shapes that yields consistent hierarchical part segmentations for downstream part-based tasks.
feature-fieldpart-basedsegmentation -
2025
Physical-grounded 3D asset generation that augments generated shapes with physical properties such as materials and kinematics so assets are simulation-ready.
asset-generationphysicssimulation -
2025
A sparse, high-resolution 3D shape representation and VAE designed to reconstruct fine geometric detail as a latent backbone for high-fidelity 3D generation.
high-resolutionrepresentationsparsevae -
Topology-Preserved Auto-regressive Mesh Generation in the Manner of Weaving Silk Discovered by AI
2025
A compact and efficient mesh representation and tokenization algorithm that generates topology-preserved polygon meshes in an autoregressive manner akin to silk weaving.
autoregressivemeshtopology -
TreeMeshGPT: Artistic Mesh Generation with Autoregressive Tree Sequencing Discovered by AI
2025
An autoregressive Transformer that generates high-quality artistic meshes aligned with input point clouds using a novel Autoregressive Tree Sequencing based on triangle adjacency.
autoregressivemeshtransformer -
2025
A 3D shape VAE built on a sparse flexible (SparseFlex) representation that reconstructs high-resolution meshes with fine detail and open surfaces.
high-resolutionrepresentationsparsevae -
Baking Gaussian Splatting into Diffusion Denoiser for Fast and Scalable Single-stage Image-to-3D Generation and Reconstruction Discovered by AI
2024
A single-stage 3D diffusion model that directly outputs 3D Gaussian point clouds at each timestep, enforcing view consistency and enabling fast, scalable object generation and scene reconstruction from a single view.
3d-generationdiffusiongaussian-splatting -
BPT (Blocked and Patchified Tokenization)
2024
Scales mesh generation via a blocked and patchified compressive tokenization that reduces sequence length, enabling generation of meshes with more faces.
autoregressivemeshtokenization -
2024
An autoregressive auto-encoder for artistic mesh generation that compresses mesh tokens to handle higher face counts and finer detail.
autoencoderautoregressivemesh -
2024
An expressive representation of general 3D shapes that models open, non-watertight surfaces by parameterizing meshes on a manifold, supporting differentiable rendering and generation.
differentiable-renderingmeshrepresentation -
2024
Unifies 3D mesh generation with large language models by representing mesh vertices and faces as text, enabling text-conditioned mesh generation within an LLM.
llmmeshtext-to-3d -
2024
Converts shapes from various 3D representations into artist-created meshes using an autoregressive transformer, producing meshes with far fewer faces.
autoregressivemeshtransformer -
2024
Improves artist-created mesh generation with an adjacent mesh tokenization scheme that increases efficiency and the number of faces that can be represented.
autoregressivemeshtokenization -
2024
Generates articulated meshes using structure-guided transformers that model object parts and their articulation jointly.
articulatedmeshtransformer -
MeshFormer Discovered by AI
2024
A sparse-view 3D reconstruction model that explicitly leverages 3D native structures, input guidance, and training supervision to generate high-quality textured meshes in seconds. It stores features in 3D sparse voxels and learns an additional normal texture to enhance geometric details.
meshreconstructionsparse-view -
Meshtron Discovered by AI
2024
An autoregressive mesh generation model capable of generating high-quality, artist-like 3D meshes with up to 64K faces at 1024-level coordinate resolution. It utilizes an Hourglass Transformer architecture and sliding window attention to achieve high efficiency and scalability.
autoregressivemeshtransformer -
2024
A generative 3D foundation model that uses a neural coordinate field to represent and autoregressively generate meshes at scale.
autoregressivefoundation-modelmesh -
Meta 3D AssetGen Discovered by AI
2024
A text-to-3D and image-to-3D generation pipeline that produces high-quality meshes with physically-based rendering (PBR) materials. It generates multi-view factored shaded and albedo appearance channels, then reconstructs colors, metalness, and roughness in 3D to support realistic relighting.
image-to-3dmeshpbrtext-to-3d -
2024
Generates generic 3D meshes by first predicting pivot vertices that guide the subsequent autoregressive generation of the full mesh.
autoregressivemeshtransformer -
3D Gaussian Splatting for Real-Time Radiance Field Rendering
2023
Represents scenes as optimized 3D Gaussians for real-time, high-quality radiance-field rendering, now a dominant representation for captured 3D assets.
gaussian-splattingreal-time -
2023
A shape representation encoding 3D geometry as a set of latent vectors, designed for diffusion-based generative modeling of neural fields.
diffusionimplicitlatent-set -
2023
A generative diffusion framework operating directly on continuous functions, applied to generating implicit 3D shape representations.
diffusionfunction-spaceimplicit -
2023
Generates 3D meshes with a diffusion model operating on a deformable tetrahedral grid parameterization of geometry.
deformable-tetrahedradiffusionmesh -
2023
Generates triangle meshes directly as sequences of tokens using a decoder-only transformer over a learned mesh vocabulary, producing compact artist-like topology.
autoregressivemeshtransformer -
Neural Wavelet-domain Diffusion
2023
Generates 3D shapes by running a diffusion model in the wavelet domain of a truncated signed distance field for efficient high-quality synthesis.
diffusionsdfwavelet -
2023
Generates high-resolution sparse voxel hierarchies with a cascaded latent diffusion model for large-scale 3D object and scene generation.
diffusionsparsevoxel -
2022
A generative model that produces textured 3D meshes with complex topology directly, aimed at game-ready asset generation.
gantextured-mesh -
Instant Neural Graphics Primitives (Instant-NGP)
2022
A multiresolution hash encoding that trains NeRFs and neural fields in seconds, a foundational speedup behind many image-to-3D reconstruction pipelines.
hash-encodingnerf -
2022
Adapts a StyleGAN-style generator to signed distance fields for high-quality 3D shape generation with global and local discriminators.
gansdfshape-generation -
2022
A latent diffusion framework over truncated signed distance fields that supports multimodal 3D shape generation and completion conditioned on images or text.
diffusionmultimodalsdf -
2021
A data-driven approach to mesh extraction from implicit fields that learns to reconstruct sharp features and correct local topology.
implicitisosurfacemesh -
2021
A probabilistic diffusion model on point-voxel representations that performs both unconditional 3D shape generation and shape completion from partial inputs.
completiondiffusionpoint-cloud -
Convolutional Occupancy Networks
2020
Combines convolutional features with occupancy prediction to represent detailed 3D geometry and reconstruct objects and scenes at scale.
implicitoccupancyreconstruction -
2020
An autoregressive generative model of 3D meshes that sequentially predicts mesh vertices and then faces using transformer-based pointer networks.
autoregressivemeshtransformer -
2020
Implicit neural representations using periodic sine activation functions that capture fine detail and derivatives, applied to signals including 3D shapes.
implicitneural-fieldrepresentation -
2019
Learns a continuous signed distance function conditioned on a latent code to represent a class of shapes, enabling shape representation, interpolation, and completion.
implicitsdfshape-representation -
2019
Represents and generates 3D shapes as sequences of parts using a sequence-to-sequence autoencoder for assembly-based shape synthesis.
part-basedsequentialshape-generation -
2019
A hierarchical graph network that generates 3D shapes as structured arrangements of parts, encoding both geometry and part relationships.
graphhierarchicalpart-based -
2018
An implicit field decoder that predicts inside/outside for query points to learn generative models of 3D shapes with smooth surfaces.
decoderimplicitshape-representation -
2018
A convolutional neural network that operates directly on mesh edges with learned pooling, providing a foundation for mesh-based geometric learning.
convolutionmeshrepresentation -
2016
An early generative adversarial network that synthesizes 3D objects as voxel grids from a probabilistic latent space.
ganshape-generationvoxel -
An open native-3D shape generation model and codebase focused on producing high-resolution geometry for downstream asset pipelines.
open-sourceshape-generation
Proprietary / Industry Tools¶
-
Tripo3D NEW Discovered by AI
2024
An end-to-end commercial 3D generation service offering fast text-to-3D and image-to-3D mesh generation, along with automated mesh retopology, rigging, and PBR texturing for game production workflows.
3d-generatorgame-readyretopologyrigging -
Meshy AI Discovered by AI
2023
A popular web-based AI 3D generation platform designed for game developers and artists to convert text prompts or 2D images into fully textured, export-ready 3D assets. It features automatic PBR texturing, remeshing, rigging, and animation support.
image-to-3dmeshpbrtext-to-3d -
Tripo AI Discovered by AI
2023
A fast AI 3D model generator that transforms text, images, or sketches into detailed 3D meshes with clean topology and rich textures in seconds. It supports quad remeshing, parts segmentation, and direct export to major 3D formats.
image-to-3dmeshtext-to-3d -
Kaedim Discovered by AI
2020
A professional AI-powered 3D asset creation platform that transforms 2D images, sketches, and reference packs into production-ready 3D meshes. It is widely used by game studios to accelerate asset pipelines while maintaining creative control.
game-assetsimage-to-3dmesh
