awesome-3D-base-models

Similar to the 2D Base Model (such as Stable Diffusion, Flux) that builds a bridge between text and images, the 3D Base Model is a bridge between text/images and 3D data. But the pipeline of 3D creation is way more complex than 2D creation.

3D Animation Pipeline: Modeling (Re-Topology / UV Unwrapping / Artist Mesh) → Texture (PBR) → Motion (Rigging & Skinning / Animation / Rendering / Lighting / Simulation)

Review

1. Feed-Forward-3D

Modeling - 3D Representation

1. VecSet (TOG 2023), Used by Hunyuan3D2.1, TripoSG, Step1X-3D)
2. (SLat) TRELLIS (Dec, 2024, CVPR'25)
3. Ghost on the Shell: An Expressive Representation of General 3D Shapes (ICLR 2024 Oral)
4. FastMesh: Efficient Artistic Mesh Generation via Component Decoupling
5. Geometry Distributions (ICCV 2025)
6. (VoxSet) LATTICE: Democratize High-Fidelity 3D Generation at Scale (Nov, 2025)
7. (O-Voxel) TRELLIS.2 (Dec, 2025)
8. Faithful Contouring (Nov, 2025)
9. FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation (CVPR 2026)

Modeling - VAE For Mesh Reconstruction

1. Sparc3D (Jun, 2025)
2. TripoSF (Mar, 2025)
3. Dora (Feb, 2025, CVPR'25)

Modeling/Texture - Flow-matching DiT For Mesh Generation & Texture Generation

1. Ultra3D (Jul, 2025)
2. Hunyuan3D-2.1 (Jun, 2025)
3. Hunyuan3D-2 (Jan, 2025)
4. TripoSG (Mar, 2025)
5. Step1X-3D (May, 2025) with 🏋️training code & 🔥dataset!
6. Direct3D-S2 (May, 2025)
7. Hi3DGen (Apr, 2025 ICCV'25)
8. TRELLIS (Dec, 2024, CVPR'25)
9. UniLat3D (Sep, 2025)
10. UltraShape 1.0
11. Trelli2

Modeling - Re-topology / Artist Mesh Generation

1. BPT (CVPR' 25)
2. DeepMesh (ICCV' 25)
3. MeshMosaic (CVPR' 26)
4. SATO (SIGGRAPH' 26)

Modeling - UV unwrapping

1. Auto-Regressive Surface Cutting (Jun, 2025)
2. ArtUV: Artist-style UV Unwrapping (Sep, 2025)
3. SeamCrafter: Enhancing Mesh Seam Generation for Artist UV Unwrapping via Reinforcement Learning (Sep, 2025)
4. PartUV: Part-Based UV Unwrapping of 3D Meshes (SIGGRAPH Asia'24)
5. MeshTailor: Cutting Seams via Generative Mesh Traversal

Textrue Generation

1. SyncMVD (SIGGRAPH Asia'24)
2. TEXGen (SIGGRAPH Asia'24)
3. ARM(CVPR'25)
4. MaterialAnything (CVPR'25)
5. MaterialMVP (ICCV'25)
6. RomanTex (ICCV'25)
7. NaTex: Meet Native Texture Generation (Nov 25)
8. SeqTex: Generate Mesh Textures in Video Sequence (SIGGRAPH Asia'25)
9. Chord: Chain of Rendering Decomposition for PBR Material Estimation from Generated Texture images (SIGGRAPH Asia'25)
10. LSRM: High-Fidelity Object-Centric Reconstruction via Scaled Context Windows

Motion - Rigging / Articulation / Dynamics with Physical Property / Animation / Simulation

1. Unirig (SIGGRAPH'25)
2. AnimateAnyMesh(ICCV'25)
3. FreeArt3D: Training-Free Articulated Object Generation using 3D Diffusion(SIGGRAPH AISA'25)
4. Pixie: Physics from Pixels(Aug 2025)
5. SAFT: Shape and Appearance of Fabrics from Template via Differentiable Physical Simulations from Monocular Video (ICCV 2025)
6. BrickGPT: Generating Physically Stable and Buildable Brick Structures from Text (ICCV 2025 Best Paper)
7. PhysX-3D: Physical-Grounded 3D Asset Generation (NIPS 2025 Spotlight)
8. PhysX-Anything: Simulation-Ready Physical 3D Assets from Single Image
9. PGC: Physics-Based Gaussian Cloth from a Single Pose (CVPR 2025 Highlight)
10. PhysAvatar: Learning the Physics of Dressed 3D Avatars from Visual Observations (ECCV 2024)
11. DSO: Aligning 3D Generators with Simulation Feedback for Physical Soundness (ICCV 2025)

Acceleration For Mesh Generation (lightning vecset decoder)

1. FlashVDM (ICCV' 25)

Post-train of 3D base models

1. DeepMesh: Auto-Regressive Artist-Mesh Creation With Reinforcement Learning (ICCV 2025)
2. Nabla-R2D3 : Effective and Efficient 3D Diffusion Alignment with 2D Rewards (NIPS 2025)
3. DreamDPO: Aligning Text-to-3D Generation with Human Preferences via Direct Preference Optimization (ICML 2025)

Common Metrics

For mesh

1. ULIP-T(⬆), ULIP-I(⬆), ULIP
2. Uni3D-T(⬆), Uni3D-I(⬆), Uni3D
   • code for ULIP/Uni3D: https://github.com/Tencent-Hunyuan/FlashVDM/blob/main/evaluation/app_metric.py
3. Chamfer Distance (CD⬇) and Normal Consistency (NC⬆) quantify the geometric similarity between reconstructed/generated and ground-truth meshes.

For texture

1. CLIP-FiD(⬇)
2. CMMD(⬇)
3. CLIP-I(⬆)
4. LPIPS(⬇)

Benchmark

1. HY3D-Bench