A self-contained, portable package for running Hunyuan3D v2 on Windows w/ a custom gradio build.
- Convert 2D images into 3D models (Base Mesh & Textured Mesh)
- Supports Text-to-3D generation (enabled by default)
- Optimized for low-VRAM GPUs w/ fallback modes
- One click finalize build – portable & easy to use
- Viewer for previous generations
- Windows 10 or 11
- NVIDIA GPU w/ at least 6GB VRAM (4GB may work but will be slow)
- CUDA Toolkit 12.8 (For texture generation)
- Visual Studio Build Tools 2022 (For compiling)
- 8GB+ VRAM for full shape + texture generation
- 24GB+ RAM for smooth processing
- Latest NVIDIA drivers (March 2024 or newer)
- Download HY3D_v2_WinPortable_v0.1.3.7z from HY3D_v2_WinPortable_v0.1.3
- Extract the .7z file to a location of your choice.
- Open the extracted folder.
- Run
run_build.bat
To enable textured model generation, install:
- CUDA Toolkit 12.8 (Download)
- Visual Studio Build Tools 2022 (Download)
- During setup, select: "Desktop Development w/ C++"
- Run
boot_HY3D_v2.bat - Access the WebUI: Open http://127.0.0.1:7860 or http://localhost:7860 in your browser.
- Run
boot_low_vram_HY3D_v2.bat - Access the WebUI: Open http://127.0.0.1:7860 or http://localhost:7860 in your browser.
- Upload an image or enter a text prompt in the WebUI.
- Click "Generate Base Mesh Only" to get a white 3D mesh.
- Click "Generate Base & Textured Meshes" (if texture generation is enabled).
- View/download the output in the interface.
Note - Depending on your hardware specs the generation time might be long, but if there is no error in terminal it is working, just give it a bit of time.
- 3D models are saved in:
HY3D_v2/Hunyuan3D-2/outputs/'%Y-%m-%d'/'%H-%M-%S'_steps-{num_steps}_cfg-{cfg_scale}_res-{octree_resolution}_seed-{seed}/
| File | Description |
|---|---|
HY3D_v2/Hunyuan3D-2/upgradio_app.py |
The Gradio WebUI app script |
HY3D_v2/boot_HY3D_v2.bat |
Starts the app for high VRAM GPUs |
HY3D_v2/boot_low_vram_HY3D_v2.bat |
Starts the app for low VRAM GPUs |
HY3D_v2/dnld_models.bat |
Downloads all required models (~39GB) |
HY3D_v2/oc_install.bat |
One-time setup for texture generation |
HY3D_v2/run_build.bat |
Runs both dnld_models.bat & oc_install.bat to install in one |
| Error | Solution |
|---|---|
CUDA out of memory |
Use boot_low_vram_HY3D_v2.bat. |
Can't load TensorRT |
Ignore – The app will fallback to CUDA. |
Texture generation not working |
Ensure CUDA & VS Build Tools are installed, then re-run oc_install.bat. |
Models not found |
Run dnld_models.bat again. |
- Open the app & go to UI Settings.
- Select a new theme.
- Restart the app.
If you need a proxy, add this to your .bat scripts:
set HTTP_PROXY=http://your-proxy.com:[port]
set HTTPS_PROXY=http://your-proxy.com:[port]-
Modified & Re-packaged by MackinationsAi (GitHub)
-
Special thanks to the researchers, developers, and all contributors of https://github.com/Tencent/Hunyuan3D-2[Hunyuan3D 2.0], & YanWenKun for laying out a lot of the ground work w/ their winportable version. Kudos to DeepBeepMeep for creating mmgp & Hunyuan3D-2GP for bringing Hunyuan3D 2.0 to less-capable GPUs.
