Two official open versions
Hunyuan3D-2 and Hunyuan3D-2.1 both ship Hugging Face Spaces and model repositories under the tencent org.
Tencent Hunyuan · Open image-to-3D
Hunyuan3D
Tencent open image-to-3D — Hunyuan3D-2 and production-oriented 2.1 with PBR
Hunyuan3D is Tencent’s open family of systems that turn images (and related inputs) into textured 3D meshes. This single landing covers both Hunyuan3D-2 and Hunyuan3D-2.1—so you can understand version roles and jump into either official Hugging Face demo.
What is Hunyuan3D?
Hunyuan3D-2 is published on Hugging Face as an image-to-3D / text-to-3D stack (model tags include arXiv:2501.12202 and arXiv:2411.02293). It is widely used in the open community as a two-stage shape + texture style pipeline with official Gradio Spaces.
Hunyuan3D-2.1 is described in “Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material” (arXiv:2506.15442). Official materials emphasize two foundations: Hunyuan3D-DiT for shape generation (flow-based diffusion + shape VAE) and Hunyuan3D-Paint for multi-view PBR materials (e.g. albedo / metallic / roughness channels), plus broader open-sourcing of weights and training code versus earlier texture RGB pipelines.
We intentionally do not split 2.0 and 2.1 into separate landings: product-wise they are one series. The playground embeds both Spaces as tabs so you can A/B the live demos.
When to use Hunyuan3D
Use Hunyuan3D when you want a mature Tencent open pipeline for image-to-3D meshes and textures—especially 2.1 if you care about production-oriented PBR materials and open training/finetune paths documented in the 2.1 technical report.
How Hunyuan3D works
STEP 1
Image (or text) conditioning
Start from a single object image in the Space UI. Model cards also tag text-to-3D capability depending on the deployed pipeline.
STEP 2
Generate shape (DiT / shape VAE)
Hunyuan3D-2.1’s shape path combines a flow-based diffusion transformer with a high-fidelity mesh autoencoder (Hunyuan3D-ShapeVAE per the report).
STEP 3
Paint materials
Texture models synthesize view-consistent appearance; 2.1’s Paint stage targets multi-channel PBR maps rather than simple RGB bake-only looks.
Key capabilities
Shape then texture
2.1 docs describe Hunyuan3D-DiT for geometry and Hunyuan3D-Paint for mesh-conditioned multi-view PBR synthesis.
PBR-focused 2.1 upgrade
The 2.1 report highlights physics-grounded PBR maps (e.g. metallic reflections) replacing earlier RGB-only texture painting approaches.
Community deployment
Diffusers-tagged weights, bilingual model cards (en/zh), and large Space traffic make Hunyuan3D a common open baseline.
Technical snapshot
Figures below come from the linked papers, model cards, or project pages—not from unbenchmarked third-party marketing claims.
- Hunyuan3D-2 papers
- arXiv:2501.12202, arXiv:2411.02293 (model tags)
- Hunyuan3D-2.1 paper
- arXiv:2506.15442
- 2.1 architecture
- Hunyuan3D-DiT (shape) + Hunyuan3D-Paint (PBR)
- HF models
- tencent/Hunyuan3D-2 , tencent/Hunyuan3D-2.1
- Demos
- spaces/tencent/Hunyuan3D-2 and Hunyuan3D-2.1
Who it is for
Game & XR asset blocking
Quick textured meshes for level blockouts, avatar props, and AR product previews.
Open-source finetuning
2.1 materials stress full weights and training code for community finetunes and industrial adaptation.
PBR look development
When metallic/roughness style maps matter more than pure silhouette reconstruction.
Versions covered on this page
Hunyuan3D-2.0
Established open 2.x demo (HF: tencent/Hunyuan3D-2). Strong community adoption for image-to-3D meshes and textures; related papers tagged on the model card (arXiv:2501.12202, 2411.02293).
Hunyuan3D-2.1
Updated open release focused on high-fidelity assets with production-oriented PBR materials (arXiv:2506.15442). Demo Space: tencent/Hunyuan3D-2.1.
Both demos are available as tabs on the playground—no separate landing pages for 2.0 vs 2.1.
Limitations & practical notes
- 2.0 and 2.1 are separate codebases/releases—not always a drop-in upgrade; follow each repo’s install guide.
- PBR quality and channel set depend on the exact checkpoint and Space build you run.
- Commercial licensing may differ from fully permissive MIT models—read Tencent’s license terms on each model card before redistribution.
Frequently asked questions
Why are 2.0 and 2.1 on one page?+
They are successive open releases in the same Hunyuan3D product line. One landing avoids duplicate SEO pages while the playground still embeds both Spaces.
What changed in Hunyuan3D-2.1?+
Per arXiv:2506.15442 and the official GitHub README, 2.1 emphasizes production-ready PBR material synthesis (Hunyuan3D-Paint) alongside shape generation (Hunyuan3D-DiT), with broader open-sourcing of training components.
Can I try both demos here?+
Yes. Open the Hunyuan3D playground and switch tabs between Hunyuan3D 2.0 and 2.1 official Spaces.
Sources & references
Landing claims are grounded in the following primary materials. Always prefer the original paper or model card for citations.
Try Hunyuan3D in the browser
Jump into the playground and generate a 3D model from your own image—no install required.