Image-to-3D¶
3D Generation
Open Source¶
-
img2threejs Discovered by AI
2026
An open-source AI agent tool that rebuilds objects from reference images as code-only, procedural, and animation-ready Three.js models. It outputs readable TypeScript and JSON instead of opaque mesh files, making it highly efficient for web-based game development.
image-to-3dprocedural-generationthreejswebgl -
2026
A 3D asset generation model from Tencent ARC in the flow-matching diffusion-transformer family for mesh and texture generation (SIGGRAPH 2026).
diffusionflow-matchingimage-to-3d -
Pixal3D Discovered by AI
2026
A pixel-aligned 3D generation framework that establishes direct pixel-to-3D correspondences by explicitly lifting 2D features into 3D space via back-projection. This approach achieves near-reconstruction-level fidelity and naturally extends to multi-view generation and object-separated 3D scene synthesis.
3d-reconstructionimage-to-3dpixel-alignedscene-synthesis -
Acc3D: Accelerating Single Image to 3D Diffusion Models via Edge Consistency Guided Score Distillation Discovered by AI
2025
A method designed to accelerate single image-to-3D diffusion models, achieving over a 20x increase in computational efficiency. It integrates an edge consistency-guided score distillation process and an adversarial augmentation strategy to derive high-quality reconstructions through few-step inferences.
3d-generationaccelerationdiffusion-modelimage-to-3d -
AR-1-to-3: Single Image to Consistent 3D Object Generation via Next-View Prediction Discovered by AI
2025
An autoregressive next-view prediction framework for 3D object generation from a single image. It generates target views progressively from near to far, significantly improving multi-view consistency and detail preservation.
3d-generationautoregressiveimage-to-3dmulti-view -
2025
A scalable native-3D diffusion model using spatial-sparse attention to generate high-resolution shapes directly from images while sharply cutting training cost.
diffusionimage-to-3dscalablesparse -
2025
Generates high-fidelity 3D geometry from images by bridging through normal maps, preserving fine surface detail that direct image-to-3D often loses.
high-fidelityimage-to-3dnormal-bridge -
Hi3DGen Discovered by AI
2025
A high-fidelity 3D geometry generation framework that reconstructs detailed shapes from single images using normal bridging. It decouples low- and high-frequency details via an image-to-normal estimator, resolving domain gaps and visual ambiguities to produce highly accurate 3D models.
geometrymeshnormal-mapreconstruction -
Hunyuan3D 2.0 Discovered by AI
2025
An advanced large-scale 3D synthesis system that generates high-resolution textured 3D assets from a single image or text. It utilizes a flow-based diffusion transformer (Hunyuan3D-DiT) for geometry generation and a texture synthesis model (Hunyuan3D-Paint) to produce detailed, high-fidelity 3D meshes.
diffusionmeshtexturetransformer -
Hunyuan3D 2.5 Discovered by AI
2025
An upgraded suite of 3D diffusion models that introduces a new shape foundation model called LATTICE, trained with scaled high-quality datasets and increased compute. It significantly advances both shape and texture generation to deliver high-fidelity 3D assets with ultimate geometric details.
diffusionhigh-fidelitylatticemesh -
2025
A large-scale open image-to-3D system pairing a shape diffusion transformer with texture synthesis to produce detailed textured meshes from a single image.
diffusionimage-to-3dopen-weightstexture -
2025
Tencent's open, production-grade image-to-3D system that generates high-resolution shapes with PBR materials through a two-stage shape-then-texture pipeline.
diffusionimage-to-3dopen-weightspbr -
HunyuanWorld 1.0: Generating Immersive, Explorable, and Interactive 3D Worlds from Words or Pixels Discovered by AI
2025
A framework for generating immersive, explorable, and interactive 3D worlds from text or images, combining video-based diversity with 3D geometric consistency.
3d-sceneimage-to-3dtext-to-3dworld-generation -
MapAnything Discovered by AI
2025
A unified transformer-based feed-forward model from Meta that directly regresses metric 3D scene geometry and camera parameters from one or more images.
3d-reconstructionfeed-forwardmetric-depthtransformer -
2025
Generates simulation-ready physical 3D assets from a single image, predicting geometry together with physical attributes for direct use in physics engines.
image-to-3dphysicssimulation -
SAM 3D Discovered by AI
2025
A generative model by Meta for visually grounded 3D object and human body reconstruction from a single image, predicting geometry, texture, and layout.
3d-reconstructionmetasam3dsingle-view -
SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images Discovered by AI
2025
SPAR3D (Stable Point-Aware Reconstruction of 3D Objects) is a single-image 3D reconstruction framework that combines regression and generative modeling. It uses a point-aware representation to efficiently reconstruct visible surfaces while leveraging generative priors to synthesize plausible and aligned geometry for occluded regions.
image-to-3dmesh-generationpoint-cloudstability-ai -
Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modeling Discovered by AI
2025
A unified 3D shape modeling framework that utilizes sparse representation to generate high-resolution 3D meshes. It addresses detail loss in traditional VAE-based pipelines by directly learning a sparse latent representation of 3D geometry.
image-to-3dmeshsparse-representationvae -
SparseFlex Discovered by AI
2025
A sparse-structured isosurface representation that enables differentiable mesh reconstruction at high resolutions directly from rendering losses. It supports creating high-fidelity 3D meshes with arbitrary topology, including open surfaces and complex interiors, without requiring costly watertight conversions.
isosurfacemeshreconstructiontopology -
2025
An open image-to-3D framework released with training code and a high-quality dataset, generating geometry and texture in two stages for asset creation.
datasetdiffusionimage-to-3dopen-weights -
2025
The successor to TRELLIS, using an object-centric voxel (O-Voxel) structured latent to generate higher-fidelity 3D assets from text or images.
diffusionimage-to-3dstructured-latentvoxel -
2025
A large rectified-flow transformer for high-fidelity image-to-3D shape generation, trained on a curated large-scale 3D dataset.
diffusionflow-matchingimage-to-3d -
TripoSG Discovered by AI
2025
A high-fidelity 3D shape synthesis framework based on large-scale rectified flow models. It addresses the limitations of 3D data scale and complexity, enabling rapid and high-quality reconstruction of detailed 3D meshes from single-view images.
meshreconstructionrectified-flow -
2025
A unified-latent approach to 3D generation that encodes shapes into a single latent space for native 3D asset synthesis from text or images.
diffusionimage-to-3dunified-latent -
Wonder3D++: Cross-domain Diffusion for High-fidelity 3D Generation from a Single Image Discovered by AI
2025
An enhanced single-image-to-3D generation framework that leverages cross-domain diffusion to generate consistent multi-view normal maps and color images. It employs a cascaded explicit-mesh reconstruction pipeline to rapidly produce high-fidelity textured 3D meshes.
diffusionimage-to-3dmeshnormal-map -
WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting Discovered by AI
2025
An all-in-one, feed-forward model for versatile 3D geometric prediction tasks that integrates diverse geometric priors and generates multiple 3D representations including dense point clouds, multi-view depth maps, and 3D Gaussians.
3d-reconstructiongaussian-splattingmulti-viewtencent -
2024
Represents scenes with 2D oriented Gaussian disks instead of 3D ellipsoids for geometrically accurate radiance fields and improved surface reconstruction.
gaussian-splattingsurface-reconstruction -
2024
Generates high-fidelity 3D meshes with a native 3D diffusion model followed by a normal-based surface refinement stage for smoother geometry.
diffusionimage-to-3dmesh -
CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Model Discovered by AI
2024
Convolutional Reconstruction Model (CRM) is a high-fidelity feed-forward single image-to-3D generative model that leverages the geometric priors of the triplane component in its architecture to generate 3D textured meshes in 10 seconds.
image-to-3dmesh-generationreconstruction -
Direct3D NEW Discovered by AI
2024
A native 3D generative model scaling image-to-3D generation using a 3D Variational Auto-Encoder and a 3D Latent Diffusion Transformer.
3d-reconstructionditimage-to-3dlatent-diffusion -
Era3D: High-Resolution Multiview Diffusion using Efficient Row-wise Attention Discovered by AI
2024
Era3D is a high-resolution multiview diffusion method that generates high-resolution multiview images and normal maps from a single-view image using efficient row-wise attention, automatically estimating camera focal length and elevation.
diffusionimage-to-3dmulti-view -
Grounding Image Matching in 3D with MASt3R Discovered by AI
2024
An extension of DUSt3R that integrates dense local feature matching with 3D reconstruction, significantly improving camera pose estimation and matching robustness under extreme viewpoint changes.
3d-reconstructionimage-matchingpose-estimation -
Hunyuan3D 1.0: A Unified Framework for Text-to-3D and Image-to-3D Generation Discovered by AI
2024
A unified framework for text-to-3D and image-to-3D generation that employs a two-stage pipeline. It first generates multi-view RGB images using a multi-view diffusion model and then reconstructs high-quality 3D meshes or Gaussian splats.
image-to-3dmeshmulti-viewtext-to-3d -
InstantMesh Discovered by AI
2024
A feed-forward framework for instant 3D mesh generation from a single image, synergizing a multiview diffusion model with a sparse-view reconstruction model based on the LRM architecture.
3d-meshfast-generationlrmtencent -
LGM Discovered by AI
2024
A novel framework designed to generate high-resolution 3D models from text prompts or single-view images using multi-view Gaussian splatting as the core representation.
gaussian-splattinghigh-resolutionsingle-view -
MeshLRM: Large Reconstruction Model for High-Quality Meshes Discovered by AI
2024
A large reconstruction model that incorporates differentiable mesh extraction and rendering to reconstruct high-quality 3D meshes from sparse images in under one second.
lrmmesh-generationreconstruction -
MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Views Discovered by AI
2024
MVSplat360 is a feed-forward approach for 360-degree novel view synthesis of diverse real-world scenes from sparse observations, combining geometry-aware 3D reconstruction with temporally consistent video generation.
gaussian-splattingimage-to-3dscene-generation -
MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images Discovered by AI
2024
MVSplat is an efficient feed-forward 3D Gaussian Splatting model that predicts clean 3D Gaussians from sparse multi-view images by building a cost volume representation via plane sweeping.
gaussian-splattingimage-to-3dreconstruction -
Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion Discovered by AI
2024
A unified 3D generation framework that integrates diffusion-based multi-view image generation and 3D reconstruction into a recursive diffusion process. It uses 3D-aware maps rendered by the reconstruction module as feedback to guide the multi-view generation, ensuring high geometric consistency.
diffusionimage-to-3dmulti-viewreconstruction -
Stable Fast 3D (SF3D) Discovered by AI
2024
A fast and high-quality textured object mesh reconstruction model that generates 3D meshes from a single image in 0.5 seconds, incorporating UV unwrapping and material parameter prediction.
3d-meshfast-generationstability-aiuv-unwrapping -
2024
A structured 3D latent representation and rectified-flow transformer that generates high-quality 3D assets in multiple output formats from text or images.
diffusionimage-to-3dstructured-latent -
TripoSR: Fast 3D Object Reconstruction from a Single Image Discovered by AI
2024
TripoSR is a fast feed-forward 3D reconstruction model that generates high-quality 3D meshes from a single image in under 0.5 seconds. Developed in collaboration between Tripo AI and Stability AI, it leverages a transformer-based architecture optimized for rapid inference and high-fidelity geometry generation.
fast-inferenceimage-to-3dmesh-generationtransformer -
Unique3D Discovered by AI
2024
An image-to-3D framework for efficiently generating high-quality 3D meshes from single-view images, featuring state-of-the-art generation fidelity and strong generalizability.
3d-meshreconstructionsingle-view -
2023
Uses generative Gaussian splatting with mesh extraction and texture refinement for efficient image-to-3D and text-to-3D content creation in minutes.
gaussian-splattingimage-to-3dtext-to-3d -
DUSt3R: Geometric 3D Vision Made Easy Discovered by AI
2023
A dense and unconstrained stereo 3D reconstruction framework that directly regresses 3D pointmaps from image pairs without requiring prior camera calibration or pose estimation.
3d-reconstructionmultiviewpointmap -
2023
Adds 3D smoothing and 2D Mip filters to 3D Gaussian splatting to eliminate aliasing artifacts across varying sampling rates and zoom levels.
anti-aliasinggaussian-splatting -
2023
Reconstructs a 3D mesh from a single image in around 45 seconds without per-shape optimization by combining multi-view synthesis with a generalizable reconstruction module.
diffusionimage-to-3dmesh -
2023
Generates 3D objects from a single image using consistent multi-view generation combined with a 3D diffusion stage for improved quality and speed.
diffusionimage-to-3dmulti-view -
2023
Aligns 3D Gaussians to scene surfaces to enable efficient and accurate mesh extraction and high-quality mesh-based rendering from Gaussian splatting.
gaussian-splattingmeshsurface-reconstruction -
Wonder3D: Single Image to 3D using Cross-Domain Diffusion Discovered by AI
2023
Wonder3D is a single-image to 3D generation method that produces high-fidelity textured meshes by generating highly consistent multi-view normal maps and color images via cross-domain diffusion.
diffusionimage-to-3dreconstruction -
2023
A diffusion model that synthesizes novel views of an object from a single image by conditioning on relative camera viewpoint, enabling zero-shot single-image 3D reconstruction.
diffusionimage-to-3dnovel-view -
2021
Reconstructs surfaces from multi-view images by representing geometry as a signed distance function rendered through unbiased volume rendering.
neural-renderingsdfsurface-reconstruction -
2018
Represents a 3D surface as a collection of learned parametric patches that are deformed to reconstruct shapes from point clouds or single images.
meshreconstructionsurface
Proprietary / Industry Tools¶
-
Hyper3D Rodin Discovered by AI
2024
A premium commercial AI 3D generator that transforms text descriptions and 2D images into hyper-realistic 3D assets with detailed geometry and 4K PBR textures.
commercial-toolimage-to-3dpbr-textures -
A commercial text/image-to-3D service producing game-ready textured meshes with PBR and auto-rigging; credit-based API and web app widely used in indie pipelines.
apiauto-rigcommercialtexture
