SupaVoxel
Open-source 3D model · Tencent

Hunyuan 3D 2.0, Explained: Models, VRAM, License

Hunyuan 3D 2.0 is Tencent's open-weight image-to-3D model from January 2025; this page covers what it is and how to run it. No GPU to spare? Try SupaVoxel's generator: a textured model in typically 3 to 5 minutes.

  • SupaVoxel: 3 free credits
  • No GPU or install
  • STL, OBJ, USDZ, 3MF on paid plans

By ·Updated

Steampunk robot knight, weathered teal metal, shield and gears
ExampleSteampunk robot knight, weathered teal metal, shield and gears
Press Generate to see it in 3D
Image → 3D

Drop in a picture, get a 3D model

Upload an image
Drag & drop or click · PNG, JPG, WebP, HEIC · up to 10 MB
Or try an example

3 free credits · Google sign-in · no card

Jan 21, 2025
Hunyuan 3D 2.0 weights and code released
1.1B + 1.3B
Shape model and paint model parameters
16 GB
VRAM Tencent lists for shape plus texture
3–5 min
Typical textured model on SupaVoxel's own generator
01

What is Hunyuan 3D 2.0?

Hunyuan 3D 2.0 is an open-weight 3D generation system from Tencent's Hunyuan team that turns one picture into a textured 3D mesh. Tencent released the inference code and pretrained weights on January 21, 2025, together with a technical report, a public Hugging Face demo and its own web studio.

It introduced the two-stage design that later versions keep. A shape model called Hunyuan3D-DiT, a flow-based diffusion transformer, predicts a bare mesh that follows your image, including the sides the photo never showed. A texture model called Hunyuan3D-Paint then paints a high-resolution color map onto that mesh. Because the two stages are separate, the paint model can also texture a mesh you modeled yourself.

Hunyuan 3D 2.0 is not the newest Hunyuan3D. Tencent followed it with 2.1 in June 2025 and moved to hosted-only 3.x models later. Our Hunyuan3D overview covers the whole family; this page sticks to the 2.0 open-source release.

02

Hunyuan3D-2 model zoo: every 2.0 checkpoint

The tencent/Hunyuan3D-2 repository on Hugging Face holds several checkpoints. These are the ones Tencent lists for the 2.0 series, plus the 2mini and 2mv sister models it published in their own repositories in March 2025.

Hunyuan3D-DiT-v2-0

What it does
Image to shape (the main model)
Released
Jan 21, 2025
Size
1.1B

Hunyuan3D-Paint-v2-0

What it does
Texture generation
Released
Jan 21, 2025
Size
1.3B

Hunyuan3D-Delight-v2-0

What it does
Removes lighting from the input image
Released
Jan 21, 2025
Size
1.3B

Hunyuan3D-DiT-v2-0-Fast

What it does
Guidance-distilled shape model, about half the DiT time
Released
Feb 3, 2025
Size
1.1B

Hunyuan3D-DiT-v2-0-Turbo

What it does
Step-distilled shape model
Released
Mar 19, 2025
Size
1.1B

Hunyuan3D-Paint-v2-0-Turbo

What it does
Distilled texture model
Released
Apr 1, 2025
Size
1.3B

Hunyuan3D-2mini / 2mv

What it does
0.6B mini shape model / multi-view shape model
Released
Mar 18, 2025
Size
0.6B / 1.1B
CheckpointWhat it doesReleasedSize
Hunyuan3D-DiT-v2-0Image to shape (the main model)Jan 21, 20251.1B
Hunyuan3D-Paint-v2-0Texture generationJan 21, 20251.3B
Hunyuan3D-Delight-v2-0Removes lighting from the input imageJan 21, 20251.3B
Hunyuan3D-DiT-v2-0-FastGuidance-distilled shape model, about half the DiT timeFeb 3, 20251.1B
Hunyuan3D-DiT-v2-0-TurboStep-distilled shape modelMar 19, 20251.1B
Hunyuan3D-Paint-v2-0-TurboDistilled texture modelApr 1, 20251.3B
Hunyuan3D-2mini / 2mv0.6B mini shape model / multi-view shape modelMar 18, 20250.6B / 1.1B

Sizes and dates are as Tencent lists them in its model zoo, checked November 2026.

03

Hunyuan 3D 2.0 specs and hardware requirements

What you need to run the open Hunyuan 3D 2.0 weights yourself, as Tencent documents them. The last row shows SupaVoxel's own generator for comparison.

Maker
Tencent Hunyuan team
Input
One image; multi-view input via the separate Hunyuan3D-2mv model
Output
A mesh you can save as GLB or OBJ; textured after the paint stage
VRAM
6 GB for shape only, 16 GB for shape plus texture
Operating systems
macOS, Windows and Linux
Ways to run it
Python code, a local Gradio app, a local API server and a Blender add-on
License
Tencent Hunyuan 3D 2.0 Community License
Alternative: SupaVoxel's generator
Runs in the browser, textured model in typically 3 to 5 minutes, downloads and STL / OBJ / USDZ / 3MF exports on paid plans
04

How to run Hunyuan 3D 2.0 locally

Tencent's README boils the setup down to these steps. Expect to compile two small extensions for the texture stage.

  1. 01

    Install PyTorch and the requirements

    Install PyTorch for your GPU, clone the Hunyuan3D-2 repository from GitHub, then install its requirements and the package itself.
  2. 02

    Build the texture extensions

    The paint stage needs a custom rasterizer and a differentiable renderer, both built from the repository with their own setup scripts.
  3. 03

    Generate a shape, then paint it

    Load the shape pipeline from Tencent's Hugging Face repository, pass it your image to get a mesh, then pass that mesh and the same image to the paint pipeline.
  4. 04

    Or use the Gradio app

    Launch the bundled Gradio app with the low-VRAM flag for a local web interface, or the turbo checkpoints with FlashVDM for faster shapes.
05

Hunyuan 3D 2.0 benchmark results

Tencent compared Hunyuan 3D 2.0 with unnamed open and closed models on four image-based metrics. Lower is better for the first three, higher for CLIP-score.

Hunyuan3D 2.0

CMMD
3.193
FID_CLIP
49.165
FID
282.429
CLIP-score
0.809

Top open-source model

CMMD
3.591
FID_CLIP
54.639
FID
289.287
CLIP-score
0.787

Best closed model on each metric

CMMD
3.218
FID_CLIP
49.744
FID
294.628
CLIP-score
0.806
ModelCMMDFID_CLIPFIDCLIP-score
Hunyuan3D 2.03.19349.165282.4290.809
Top open-source model3.59154.639289.2870.787
Best closed model on each metric3.21849.744294.6280.806

These are Tencent's own numbers; the baselines are not named and we have not reproduced them. The closed row takes the best of Tencent's three closed baselines for each metric.

06

Why people still use the 2.0 release

Newer versions exist, but the 2.0 release stays popular for a few practical reasons.

Fits a consumer GPU

Tencent lists 16 GB of VRAM for shape plus texture, against 29 GB for the 2.1 release. A shape-only run needs about 6 GB.

Shape only when you want it

The shape model runs on its own, so you can make an untextured mesh for printing and skip the paint stage entirely.

Textures your own meshes

Hunyuan3D-Paint accepts a handmade mesh plus a reference image, which makes it useful even if you never use the shape model.

Fast and mini variants

Fast and Turbo shape models cut generation time, and the 0.6B 2mini model is lighter still.

Large tool ecosystem

Community ComfyUI nodes, a Windows portable build and Tencent's Blender add-on all target this release.

Widely tested

Its GitHub repository passed 15,000 stars by November 2026, so most problems already have a public answer.
07

Hunyuan 3D 2.0 limits to plan for

Running the model yourself is free, but these are the costs and limits to know before you start.

Do
  • Use a GPU with 16 GB of VRAM if you want textures
  • Start with the Gradio app's low-VRAM mode
  • Feed one object per image on a plain background
  • Read the license before you ship anything commercial
Avoid
  • Expect PBR materials: those arrived with 2.1
  • Use it in the EU, the UK or South Korea; the license excludes them
  • Use its outputs to train or improve another AI model
  • Skip the texture extensions build; painting fails without it
08

Hunyuan 3D 2.0 vs 2.1 and 3.x

The quick version: 2.0 is the lighter open model, 2.1 is the heavier open model with PBR, and 3.x is hosted only.

Released

Hunyuan3D 2.0
Jan 2025
Hunyuan3D 2.1
Jun 2025
Hunyuan3D 3.x
Sep 2025 onward

Weights

Hunyuan3D 2.0
Open
Hunyuan3D 2.1
Open, with training code
Hunyuan3D 3.x
Hosted only

Shape / paint size

Hunyuan3D 2.0
1.1B / 1.3B
Hunyuan3D 2.1
3.3B / 2B
Hunyuan3D 3.x
Not listed

VRAM, shape + texture

Hunyuan3D 2.0
16 GB
Hunyuan3D 2.1
29 GB
Hunyuan3D 3.x
Not applicable (hosted by Tencent)

Materials

Hunyuan3D 2.0
Color texture
Hunyuan3D 2.1
PBR
Hunyuan3D 3.x
Color, optional PBR
Hunyuan3D 2.0Hunyuan3D 2.1Hunyuan3D 3.x
ReleasedJan 2025Jun 2025Sep 2025 onward
WeightsOpenOpen, with training codeHosted only
Shape / paint size1.1B / 1.3B3.3B / 2BNot listed
VRAM, shape + texture16 GB29 GBNot applicable (hosted by Tencent)
MaterialsColor texturePBRColor, optional PBR

More on the whole family on the Hunyuan3D page, and how it stacks up against Microsoft's model in Hunyuan3D vs TRELLIS 2.

09

Try Hunyuan 3D 2.0 online, or skip the GPU

You can try Hunyuan 3D 2.0 without installing anything on Tencent's free Hugging Face space; Tencent's own Hunyuan3D studio now serves its newer 3.x models. Self-hosting gives you full control but needs a 16 GB GPU and a working build of the texture extensions.

If you just want a textured model from a picture, SupaVoxel's own image to 3D generator runs online with nothing to install. Upload one image, add up to 4 extra angles if the back matters, and get a textured model in typically 3 to 5 minutes, or an untextured one usually in under a minute. New accounts get 3 free credits with no card, and failed generations refund their credits.

View your model free in the browser; downloads, including STL, OBJ, USDZ and multi-color 3MF exports, come with a paid plan. Whatever made your mesh, the free STL repair tool fixes print errors, and the mesh simplifier slims a heavy mesh for a game engine.

10

What SupaVoxel's generator makes from one picture

These six models come from SupaVoxel's gallery, each made from one image with SupaVoxel's own generator, so you can judge the result.

SupaVoxel gallery: steampunk robot knight with shield and gears
SupaVoxel gallery: bronze tiger sculpture with deep spirals
SupaVoxel gallery: chibi dragon warrior in a plush hood
SupaVoxel gallery: fantasy mushroom cottage with coral details
SupaVoxel gallery: ornate jewelry box with an enamel rose
SupaVoxel gallery: opera actress bust with dense ornament
FAQ

Frequently asked questions

What is Hunyuan 3D 2.0?

Hunyuan 3D 2.0 is Tencent's open-weight image-to-3D system, released on January 21, 2025. It works in two stages: Hunyuan3D-DiT, a 1.1B-parameter diffusion transformer, builds the shape from your image, then Hunyuan3D-Paint, a 1.3B texture model, paints a color map onto it. Code and weights are on GitHub and Hugging Face under Tencent's community license.

Is Hunyuan3D 2.0 free?

Yes, the weights and code are free to download, and Tencent's Hugging Face space lets you try it in the browser. Running it yourself costs GPU time: about 6 GB of VRAM for shape alone and 16 GB with textures. If you want a free start without a GPU, SupaVoxel's generator gives new accounts 3 free credits with no card, and you can view the result in the browser.

How much VRAM does Hunyuan3D-2 need?

Tencent lists 6 GB of VRAM for shape generation and 16 GB for shape plus texture in total. The Gradio app also has a low-VRAM mode. For comparison, the later Hunyuan3D 2.1 release needs 10 GB for shape and 29 GB for shape plus texture, so a 16 GB card can run 2.0 with textures but not 2.1.

What is the difference between Hunyuan3D 2.0 and 2.1?

Hunyuan3D 2.1, released in June 2025, uses a larger 3.3B shape model and a 2B paint model that outputs PBR materials, and Tencent released its training code too. Version 2.0 is smaller, gives a plain color texture and needs about 16 GB of VRAM instead of 29 GB. Choose 2.1 for materials, 2.0 for lighter hardware.

Can I use Hunyuan 3D 2.0 commercially?

Its license allows commercial use with conditions. It does not apply in the European Union, the United Kingdom or South Korea; if your products or services had more than 1 million monthly active users in the month before the 2.0 release date, you must request a license from Tencent; and outputs may not be used to improve other AI models. Tencent claims no rights in the outputs you generate. Read the full license before you ship.

Can I try Hunyuan3D-2 online?

Yes. Tencent runs a free Hugging Face space named tencent/Hunyuan3D-2, and its own Hunyuan3D web studio, which today serves newer models. If you want another online option, SupaVoxel's image-to-3D generator takes one picture, plus up to 4 extra angles, and returns a textured model in typically 3 to 5 minutes.

Does Hunyuan3D 2.0 support text to 3D?

The shape and paint models are conditioned on an image, not a prompt. Tencent's local API server does advertise both image and text to 3D, but the core 2.0 checkpoints are image-to-3D, so the usual route is to make a picture first. If you only have an idea in words, SupaVoxel's text to 3D page starts from a prompt.

Is Hunyuan 3D 2.0 good for 3D printing?

It can be. The shape model produces a mesh you can export to OBJ or GLB, and skipping the paint stage keeps the run to about 6 GB of VRAM. Check the mesh in a slicer and run it through the free STL repair tool if it reports holes. SupaVoxel's generator exports STL or 3MF directly on a paid plan.
Start free

No 16 GB GPU for Hunyuan 3D 2.0? Make a 3D model online

Upload one picture and get a textured 3D model in typically 3 to 5 minutes, with nothing to install. Start with 3 free credits, no card needed.

No credit card required · 3 free credits to start

Hunyuan 3D 2.0: Tencent's Open-Source 3D Model | SupaVoxel