SupaVoxel
3D AI model · Tencent

Hunyuan3D 3.1, Explained: Versions, Specs, License

Hunyuan3D is Tencent's image-to-3D model family; this page covers what it is and how it works. Want a model from your own picture now? Try SupaVoxel's generator: a textured model in typically 3 to 5 minutes.

  • SupaVoxel: 3 free credits
  • Textured in 3–5 min
  • GLB free · STL, OBJ, 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

3.1
Newest Hunyuan3D version, hosted by Tencent
1.5M
Max faces Tencent's own Hunyuan3D API allows
5,000 px
Largest input side Tencent's API accepts
3–5 min
Typical textured model on SupaVoxel's own generator
01

What is Hunyuan3D?

Hunyuan3D is a 3D generation model family built by Tencent's Hunyuan team, the same group behind the Hunyuan language and image models. You give Hunyuan3D a picture (or, on Tencent's own site, a text prompt) and it returns a textured mesh: a closed 3D shape plus a color texture painted from your image.

Every Hunyuan3D release works in two stages. A shape model, a diffusion transformer, predicts the full geometry, including the sides your photo never showed. A separate paint model then projects color and material onto that surface. Splitting the job this way is why Hunyuan3D shapes and textures can be improved and swapped independently, and why Tencent can sell a cheaper untextured "geometry" mode through its API.

The early versions (2.0 and 2.1) were released as open weights on GitHub and Hugging Face, which made Hunyuan 3D one of the most tested models in the open-source community. The 3.x generation is a step up in detail and is offered as a hosted service: Tencent's own studio and the Tencent Cloud API. You can't download the 3.x Hunyuan3D weights; you reach them through Tencent's studio or API.

02

Hunyuan3D versions and release dates

The model has moved fast: from 2.0 to 3.0 took under nine months.

Hunyuan3D 2.0

Released
Jan 21, 2025
What changed
Two-stage shape + paint pipeline; 1.1B-parameter shape model, 1.3B texture model
Weights
Open

2mini / 2mv / Turbo

Released
Mar 18–19, 2025
What changed
0.6B mini model, multi-view input, faster distilled variants
Weights
Open

Hunyuan3D 2.1

Released
Jun 13, 2025
What changed
PBR materials, full training code; 3.3B shape model and 2B paint model
Weights
Open

Hunyuan3D 2.5

Released
Jun 2025 (report)
What changed
Higher-detail shapes, shown in a technical report
Weights
Hosted only

Hunyuan3D 3.0

Released
Sep 16, 2025
What changed
Announced with 3× the modeling precision of 2.x and up to 1536³ resolution
Weights
Hosted only

Hunyuan3D 3.1

Released
2026
What changed
Sharper textures and geometry, more reference views
Weights
Hosted only
VersionReleasedWhat changedWeights
Hunyuan3D 2.0Jan 21, 2025Two-stage shape + paint pipeline; 1.1B-parameter shape model, 1.3B texture modelOpen
2mini / 2mv / TurboMar 18–19, 20250.6B mini model, multi-view input, faster distilled variantsOpen
Hunyuan3D 2.1Jun 13, 2025PBR materials, full training code; 3.3B shape model and 2B paint modelOpen
Hunyuan3D 2.5Jun 2025 (report)Higher-detail shapes, shown in a technical reportHosted only
Hunyuan3D 3.0Sep 16, 2025Announced with 3× the modeling precision of 2.x and up to 1536³ resolutionHosted only
Hunyuan3D 3.12026Sharper textures and geometry, more reference viewsHosted only

The 3.0 precision and resolution figures are Tencent's; we have not measured them.

03

Hunyuan3D specs for version 3.1: inputs, outputs and limits

These are the Hunyuan3D 3.x API limits as of October 2026. The last row shows SupaVoxel's own generator for comparison.

Maker
Tencent Hunyuan team
Input
One image (JPG, PNG or WebP), or several views of the same object; text prompts on Tencent's API
Image size on Tencent's API
128 to 5,000 px per side, up to 8 MB
Face count on Tencent's API
3,000 to 1,500,000 faces, default 500,000
Texture
Color texture, optional PBR maps (metallic, roughness, normal)
Modes on 3.1
Textured, geometry only; low-poly and sketch modes stay on 3.0
License
3.x is a hosted service; 2.x open weights use the Tencent Hunyuan Community License
Alternative: SupaVoxel's generator
Textured model in typically 3 to 5 minutes, GLB free, STL / OBJ / USDZ / 3MF on paid plans
04

Where Hunyuan3D is strongest

These are the jobs where Hunyuan3D results tend to need the least cleanup.

Faithful to the picture

Hunyuan3D sticks closely to the silhouette and proportions of your image. Characters keep their pose and product shots keep their exact outline, which matters when the model has to match a reference.

Fine surface detail

Carved patterns, armor plates, folds in fabric and small relief come through as real geometry instead of being only painted on, which is what you want for a figurine print.

Clean, even textures

The paint stage removes most baked-in lighting from your photo, so the texture looks right under new lights in Blender, Unity or a web viewer.

Solid, closed shapes

Outputs are closed volumes with few holes, so a Hunyuan3D mesh usually needs little repair before slicing.

Multi-view aware

In version 3.1, Hunyuan3D can combine several views of one object, which helps it get the back right when the front alone is ambiguous.

Stylized and realistic

It handles both chibi characters and photographic products well, a range not every 3D model covers.
05

Hunyuan3D weaknesses and how to avoid them

No 3D model is perfect. These are the most common Hunyuan3D failure cases, and the input habits that fix them.

Do
  • Use a three-quarter view so Hunyuan3D sees depth
  • Keep one object per image, filling at least half the frame
  • Upload at least 512 px on the short side
  • Use a plain or simple background
Avoid
  • Tiny thumbnails: Tencent's API rejects images under 128 px
  • Thin wires, hair strands or glass, which come out thick or solid
  • Text or logos as the subject; lettering rarely survives
  • Scenes with several objects, which merge into one blob
06

How to use Hunyuan3D, and an alternative

There are two ways to use Hunyuan3D. The 3.x models run on Tencent's own 3D Hunyuan web studio and the Tencent Cloud API, where you choose the model version and mode directly. The open 2.x weights can be self-hosted from GitHub or Hugging Face, which needs a GPU with roughly 29 GB of memory to run shape and texture together. If you specifically need a Hunyuan3D file, those are the routes to take.

If you just want a textured model from a picture, SupaVoxel's own image-to-3D generator is a separate alternative. 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, failed generations refund their credits, and the result goes straight into a browser viewer.

From there you can download the GLB for free, or on a paid plan export STL, OBJ, USDZ or a multi-color 3MF. Whichever tool made your mesh, Hunyuan3D included, the free STL repair tool fixes print errors and the mesh simplifier slims a dense mesh that is too heavy for a game engine.

07

Hunyuan3D 3.1 examples

The first four models are Hunyuan3D 3.1 results, each from one picture. The last two come from SupaVoxel's gallery, made with SupaVoxel's own generator, for comparison.

Hunyuan3D: armored samurai figurine, layered plates kept as geometry
Hunyuan3D: cast-iron teapot with its hobnail pattern
Hunyuan3D: carved wooden owl, feather relief that prints well
Hunyuan3D: vintage film camera with crisp edges and dials
SupaVoxel gallery: opera actress bust with dense ornament
SupaVoxel gallery: bronze tiger sculpture with deep spirals
08

Hunyuan3D vs TRELLIS 2, Tripo and Meshy

A quick comparison of four of the best-known image-to-3D models. Details for each are on their own pages.

Made by

Hunyuan3D 3.1
Tencent
TRELLIS 2
Microsoft
Tripo
VAST (Tripo AI)
Meshy
Meshy

Open weights

Hunyuan3D 3.1
2.x only
TRELLIS 2
Yes, MIT
Tripo
Older research models
Meshy
No

Best at

Hunyuan3D 3.1
Faithful detail, figurines
TRELLIS 2
Thin and complex shapes
Tripo
Clean game-ready meshes
Meshy
Print-ready, all-round

Text input

Hunyuan3D 3.1
Yes (Tencent API)
TRELLIS 2
No
Tripo
Yes
Meshy
Yes

Extras on its own site

Hunyuan3D 3.1
Low-poly, parts, rigging API
TRELLIS 2
Self-hosted only
Tripo
Rigging, quad mesh, segmentation
Meshy
Rigging, animation, retexture
Hunyuan3D 3.1TRELLIS 2TripoMeshy
Made byTencentMicrosoftVAST (Tripo AI)Meshy
Open weights2.x onlyYes, MITOlder research modelsNo
Best atFaithful detail, figurinesThin and complex shapesClean game-ready meshesPrint-ready, all-round
Text inputYes (Tencent API)NoYesYes
Extras on its own siteLow-poly, parts, rigging APISelf-hosted onlyRigging, quad mesh, segmentationRigging, animation, retexture

Read more: TRELLIS 2, Tripo AI, Meshy AI, or the head-to-head Hunyuan3D vs TRELLIS 2.

FAQ

Frequently asked questions

What is Hunyuan3D?

Hunyuan3D is Tencent's family of AI models that turn an image, or a text prompt on Tencent's own service, into a textured 3D mesh. It works in two stages: a shape model predicts the full geometry, then a paint model adds color and material. Versions 2.0 and 2.1 are open weights; 3.0 and 3.1 are hosted services.

Is Hunyuan3D free?

The open Hunyuan3D 2.1 weights are free to download, but you need a GPU with roughly 29 GB of memory to run shape and texture together. The 3.x Hunyuan3D models are hosted by Tencent. If you want a free start without a GPU, SupaVoxel's generator gives 3 free credits at sign-up; textured models cost 3 credits on a paid plan, and the GLB download is free.

What is new in Hunyuan 3D 3.1?

Compared with 3.0, Hunyuan 3D 3.1 improves texture sharpness and geometric accuracy and accepts more reference views of the same object for better reconstruction. On Tencent's API the low-poly and sketch modes are still only on 3.0, so 3.1 is the choice for maximum detail rather than lightweight meshes.

Is there a Hunyuan3D alternative that runs online?

Yes. Besides Tencent's own studio, SupaVoxel's image-to-3D generator is an independent online alternative: upload one picture, optionally with up to 4 extra angles, and get a textured model in typically 3 to 5 minutes. You start with 3 free credits and no card, and failed generations refund their credits automatically.

Can I use Hunyuan3D models commercially?

If you self-host the open Hunyuan3D 2.x weights, the Tencent Hunyuan Community License applies: it does not cover the EU, the UK or South Korea, and services above 1 million monthly users need a separate license from Tencent. For the hosted 3.x models, check Tencent's terms. Models made with SupaVoxel's own generator on a paid plan come with a commercial license.

Is Hunyuan3D good for 3D printing?

Yes, it is one of the better models for printing because it produces closed shapes with real surface detail. If your slicer reports errors on a Hunyuan3D mesh, run it through the free STL repair tool first. If you'd rather generate a print file in one place, SupaVoxel's generator exports STL or 3MF on a paid plan.

How long does a Hunyuan3D generation take?

Self-hosting the open Hunyuan3D 2.1 weights depends entirely on your GPU, and needs about 29 GB of memory for shape plus texture. For comparison, SupaVoxel's own generator returns a textured model in typically 3 to 5 minutes and an untextured one usually in under a minute.

Hunyuan3D or TRELLIS 2: which is better?

Neither wins everywhere. Hunyuan3D tends to follow the input picture more faithfully and gives solid, printable shapes; TRELLIS 2 handles thin parts and open surfaces like leaves or cloth better. See Hunyuan3D vs TRELLIS 2 for side-by-side examples.
Start free

Try SupaVoxel's generator on your own picture

Read up on Hunyuan3D, then see what SupaVoxel makes from one image: a textured 3D model in typically 3 to 5 minutes. Start with 3 free credits, no card needed.

No credit card required · 3 free credits to start