Hunyuan3D 2.1: open-source image to 3D
Hunyuan3D is Tencent's open image-to-3D model. It works in two stages: a diffusion transformer builds the shape, then a second model paints a PBR texture onto it. Version 2.1 is among the strongest open models for clean, detailed geometry. Its licence is not open source in the usual sense: it does not apply in the EU, the UK or South Korea.
| Made by | Tencent |
|---|---|
| Released | June 2025 (2.0 in January 2025) |
| Licence | Tencent Hunyuan Community Licence |
| Commercial use | Yes, outside the EU, UK and South Korea, and below 1 million monthly active users |
| GPU memory | 10 GB for shape, 29 GB with texture |
| Output | Mesh with PBR texture |
What Hunyuan3D does
You give it one image of an object and it returns a textured 3D mesh. Shape and texture are separate models. Hunyuan3D-DiT generates the geometry, and Hunyuan3D-Paint generates a PBR texture for it, with base colour, metallic and roughness maps. The split means you can run shape alone when you only need geometry, for example for 3D printing.
Tencent has released several open versions. 2.0 came in January 2025, with a smaller 2mini and a multi-view 2mv variant. 2.1 followed in June 2025 with better geometry and PBR textures. Tencent's newer versions run on its own service, and their weights are not public.
What you need to run it
Tencent's README gives the memory: about 10 GB of GPU memory for shape generation and 29 GB for shape and texture together, with a low-memory mode that trades speed for memory. In practice that means an RTX 4090 or 5090 for shape, and a 32 GB card or a cloud A100 for the full pipeline. It runs on Linux with CUDA. It also needs Python, PyTorch and a few compiled extensions for the texture stage, which is where most installs fail.
- Try it without installing: the official Hugging Face Space runs the full model in a browser, with a queue.
- Run it locally: clone the GitHub repo, install the requirements, and start the Gradio app.
- Use it in a pipeline: the shape and paint stages can be called from Python, and there are ComfyUI nodes.
The licence, read carefully
Hunyuan3D uses the Tencent Hunyuan Community Licence, not MIT or Apache. Three terms matter. The licence does not apply in the European Union, the United Kingdom or South Korea, so it grants no rights there. A product with more than 1 million monthly active users needs a separate licence from Tencent. And there is an acceptable use policy. Outside those limits, you can use the models and their outputs commercially.
No GPU, no setup?
2dto3D runs a production pipeline for product photos: it generates the hidden views first, then builds a textured model at real size, with AR, an embed snippet and export to 7 formats. About 5 minutes per model, from $2, nothing to install.
Your first model is $2, refunded in full within 14 days if you do not like it. Try it with a sample photo
Frequently asked
Is Hunyuan3D free?
The weights are free to download and use under the Tencent Hunyuan Community Licence. You pay for the GPU you run it on. The licence does not apply in the EU, UK or South Korea, and large products need a separate licence.
Can I use Hunyuan3D models commercially?
Yes, outside the EU, UK and South Korea, as long as your product has fewer than 1 million monthly active users and you follow the acceptable use policy. Read the licence file in the repository before you ship.
How much VRAM does Hunyuan3D need?
Version 2.1 needs about 10 GB for shape and 29 GB for shape plus texture. Version 2.0 needs about 6 GB for shape and 16 GB in total. A low-memory mode reduces the peak at the cost of speed.
Is there an online Hunyuan3D demo?
Yes. Tencent runs official demos as Hugging Face Spaces for 2.0 and 2.1. They are free and run in a queue, so a busy day means a wait.
Other open models and datasets
Facts checked 2026-11-06 against the model cards, licence files and READMEs.