This project collects some good and useful 3D vision projects in every week. Especially focused on Text-to-3D and Image-to-3D.
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| LATTICE | 3DAIGC / Image to 3D Geometry |  |
Tencent | 1. Foundation model behind Hunyuan3D-2.5 |
| DreamCube | 3D Panorama Generation |  |
HKU & Tencent |  |
| Assembler | Scalable 3D Part Assembly |  |
Tencent | 1. Part Assembly (Require Part Mesh as Input) |
| Squeeze3D: Your 3D Generation Model is Secretly an Extreme Neural Compressor | 3D Mesh Compression |  |
University of Toronto |  |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| Diffusion-Renderer | Neural Inverse Rendering Video Diffusion Models |
 |
NVIDIA | CVPR 2025 Oral |
| Hunyuan3D-2.1 | Image-to-3D |  |
Tencent | SOTA 3DAIGC Model; PBR Material |
| Sparc3D | Sparse 3D Representaion |  |
NTU | 1. Sparcubes & SparconvVAE 2. Sparubes converts Raw meshes to Sparse voxels, following the stream of XCube/TRELLIS but with 3D supervision. 3. High Resolution & Intricate Detail Encoding/Decoding, facilitating downstream 3D Generation task |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| AnimateAnyMesh | General Mesh Animation |  |
HUST & DAMO Acadamy | The first feed-forward universal mesh animation framework that enables efficient motion generation for arbitrary 3D meshes |
| PartPacker | Image-to-3D Structured Mesh Generation |
 |
NVIDIA | Structured Mesh Generation from a Single RGB Image |
| OrientationMatters | Making 3D Generative Models Orientation-Aligned |  |
ZJU & Ant | 1. Introducing Objaverse-OA, a dataset of 14,832 orientation-aligned 3D models 2. Producing models with consistent orientations |
| PartCrafter | Image-to-3D Structured Mesh Generation |
 |
PKU & ByteDance |  |
| MeshArt | Articulated Mesh Generation |  |
TUM & Meta |  |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| ShapeLLM-Omni | Unified 3D Generation and Understanding |  |
Tsinghua & PKU |  |
| UniTex | Mesh Texturing |  |
LightIllusion |  |
| NuiScene | Unbounded Outdoor Scene Generation |  |
Simon Fraser University |  |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| VG-MLLM: Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors | MLLM with 3D |  |
CUHK | VG-LLM integrates a 3D visual geometry encoder (based onΒ VGGT) with a conventional 2D visual encoder. |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| Step1X-3D | 3DAIGC / Image-to-3D |  |
StepFun(ιΆθ·ζθΎ°) | 1. Support multiple conditions 2. Beats TRELLIS and Hunyuan3D, on par with Tripov2.5 |
| AnyMate | Automatic Rigging |  |
Stanford | 1. 230K 3D Models with Rig & Skin Weights 2. Model trained on the dataset |
| Articulate-Anymesh | Open-Vocabulary 3D Articulated Objects Modeling |  |
UMASS | An automated framework that is able to convert any rigid 3D mesh into its articulated counterpart in an open-vocabulary manner. |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| PrimitiveAnything | Mesh -> Primitive Assembly Generation |  |
Tencent AIPD | 1. SIGGRAPH 2025 2. Applications in UGC (Games) 3. Encoder + Transformer + Cascaded Decoder 4. Mesh/PointCloud as input |
| LegoGPT | Pphysically stable LEGO brick models from text prompts |  |
CMU | 1. Fintuned from Llama-3.2-1B-Instruct 2. Trials and stability checks |
| Beta-Splatting | 3D GS Representation |  |
USC | 1. SIGGRAPH 2025 2. A deformable and compact approach that enhances both geometry and color representation |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| Difflocks | 3D Hair Generation |  |
Meshcapade | CVPR 2025 |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| WorldGen | Text/Image based 3D World Generation |  |
Ziyang Xie @ UIUC | Fast/360Β°/flexible rendering |
| TetWeave | Differentiable ISO-Surface Representation |  |
ETH | 1. SIGGRAPH 2025 2. Alternative of DMTet and Flexicubes 3. Intersection-free & Manifold 4. Mesh Compression & Multiview-Reconstruction etc. |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| PartField | Mesh Segmentation |  |
Nvidia | 1. Concurrent work to holopart, BUT NOT add new patches 2. High accuracy and fast inference 3. Can support a number of downstream tasks based on extracted feature fields |
| OctGPT | 3D Shape Generation |  |
Peking University | SIGGRAPH 2025 |
| DNF-Avatar | Realtime Animatable Avatar Relighting |  |
VGG Group | Arxiv |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| SyncCity | Layout/Prompts to isometric 3D worlds |  |
VGG Group | π₯1. Built upon FLUX and trellis, require 40GB+ VRAM and 10+ minutes to generate a 3x3 scene 2. Flux inpainting, scene alignment, gaussian blended and decoded with trellis decoder |
| HoloPart | Amodal Mesh Segmentation and Completion |  |
VAST-AI | 1. Use SAMesh/SAMPart3D to segment a mesh, then holopart use DiT to completes every parts 2. Support future material-editing / geometry-editing |
| UniRig | Auto-Rigging |  |
VAST-AI | 1. SOTA rigging method, automatically rig characters/animals/objects in seconds. 2. A concurrent work with MagicArticulate. |
| Hi3DGen | Image to High Quality 3D Geometry |  |
StableX | 1. Built upon trellis, using normal to generate very high detailed geometries 2. Seems that official repo NOT considered the texture generation, NEED to resort to other solutions. 3. It's ComfyUI version (Recommended using ComfyUI for trellis and Hi3DGen) |
| Objaverse++ | 3D OBject Dataset |  |
CMU | Labels 3D modeling objects in Objaverse in terms of quality score and other important traits for machine learning researchers. |
| LAM | Single Image to 3D Avatar: Large Avatar Model for One-shot Animatable Gaussian Head |  |
TongYi | 1. Renders fast on various platforms 2. Seems quality NOT that good as LHM |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| TripoSG | Image-to-3D High-Fidelity 3D Shape Synthesis using Large-Scale Rectified Flow Models |
 |
VAST-AI | 1. Rectified Flow Transformer 2. Similar to trellis |
| TripoSF | PointCloud-to-3D SparseFlex: High-Resolution and Arbitrary-Topology 3D Shape Modeling |  |
VAST-AI | 1. Integrate Sparse Voxel/Conv op with Flexicubes 2. Currently ONLY open source the reconstruction from point cloud 3. Generation ability is hard to surpass trellis (also what's it built that part upon) |
| NeuralLightRig | Relightable 3D object Reconstruction |  |
Shanghai AI Lab & CUHK | Unlocking Accurate Object Normal and Material Estimation with Multi-Light Diffusion |
| HSMR | Reconstructing Humans with a Biomechanically Accurate Skeleton |  |
ZJU | CVPR25 Oral |
| Repo/Project | Domain | Stars | Author | Notes |
|---|---|---|---|---|
| DI-PCG | Text-to-3D |  |
TencentARC | 1. Predefined procedural generators are REQUIRED 2. Similar to primitive assembly |
| Vibe-draw | Sketch-to-3D |  |
martin226 | π₯1. A try-out version: demo 2. Won the 1st place in an AI hackerthon |
| tripo-mcp | Text-image-to-3D Blender, MCP |
 |
VAST-AI | Integrate tripo 3D generator with MCP, built upon Blender-MCP |
| blender-mcp | Blender Integration |  |
ahujasid | π₯ MCP for blender, connects llm (cursor/claude code) with blender, make creative things automatically |
| LHM | Single Image to Large Animatable Human Reconstruction |  |
TongYi | Large Animatable Human Reconstruction |
| VGGT | Fundamental 3D Vision |  |
Meta | π₯1. Single Image to Point Cloud/Maps/Depth maps 2. Can be used as backbone for various downstream tasks |
| DeepMesh | AutoRegressive Mesh Generation |  |
Tsinghua & S-Lab | Autoregressive Mesh Generation with RL |
| TreeMeshGPT | AutoRegressive Mesh Generation |  |
Sea AI Lab | Autoregressive mesh Generation |
| FlashVDM | FAST Image-to-3D Geometry Generation |  |
Tencent | π₯1. Very fast (~0.5s) geometry inference via progressive Distillation and engineering optimization, upon Hunyuan3D-2 2. Time cost above DOES NOT include decimation and texture generation/baking |
| Hunyuan3D-2 | Image-to-3D |  |
Tencent | 1. High Quality 3D geometry 2. Texture a little bit over-saturated |
| Kiss3DGen | 3D generation |  |
EnVision Research | Repurposing Image Diffusion Models for 3D Asset Generation (CVPR'25) |
| MIDI-3D | Image to (Multiple) 3D |  |
VAST-AI | Multi-Instance Diffusion for Single Image to 3D Scene Generation |
| MagicArticulate | Auto Rig/Skin |  |
Seed3D | Auto Rigging and Skinning of objects from many categories |
| CanonObjaverseDataset | 3D Dataset |  |
SU & PKU | 1. For Canonicalization of 3D generative models 2. CVPR 2025 Highlight |
| Provider | Url | Models | Comments |
|---|---|---|---|
| Meshy | https://www.meshy.ai/ | Meshy1-4 | Need subscription |
| Rodin/Hyper3D | [Rodin - Free AI 3D Model Generator from Images | Hyper3D] | Rodin Gen1/1.5 | Need subscription |
| Tripo | Tripo3d.ai | Tripo1.4/2.0/2.5 | Need subscription |
| Stability-AI | Stability AI | sf3d/spar3d | ~0.02/0.04$ per 3D generation, fast inference |
| Replicate | Dashboard βΒ Replicate | trellis | ~0.07$ per 3D generation |
| fal.ai | fal.ai | The generative media platform for developers | trellis | - |
| piapi | PiAPI - central hub for your generative AI APIs | trellis | ~0.035$ per 3D generation, $0.25 credits once registered |