Hi3D v1.0: 3D Generator — What It Does and What You Need to Run It

Hi3D v1.0 is an open source 3D generator that handles the image-to-3D task: an image goes in, 3D geometry comes out. The developer positions the project as a tool for high-detail geometry refinement built on top of Sparc3D. The repository publishes the code to run it, the model weights and documentation.
What it does
The authors describe Hi3D v1.0 as a system for high-detail refinement of 3D geometry based on Sparc3D. The task the tool handles is image-to-3D, that is, converting a flat image into a volumetric model. The repository description mentions inference code, model weights and documentation. The repository tags point to 3D generation, geometry refinement, image-to-3D and a connection to Sparc3D. The authors do not specify which input image and output model formats are supported.
What you need to run it
The code is written in Python. As described by the author, running it requires a Python 3.10 environment. The platform the project runs on is CUDA. The description states that the constraints pin the measured kernel stack, including spconv-cu120 version 2.3.6 and flash-attn version 2.7.4.post1. Flash-attn is installed separately after PyTorch: building it from source requires an installed torch and the CUDA toolkit, with build isolation disabled. The Triton mesh kernels are compiled on first use. The project is installed from source; there is no ready-made package on PyPI or prebuilt wheel file.
The largest weights file is 3.4 GB, and all weights files together are 7.2 GB. The repository was created on October 16, 2026, and the last code change was on October 23, 2026. The latest release is v1.0.
The code and weights are distributed under the Hi3D Open License 1.0. The terms are the same for code and weights: you may use, modify, distribute and sell the materials, including commercial use. If you exceed the threshold of 200,000 MAU or a certain level of monthly revenue, you must notify the licensor and obtain a separate written license. Commercial products must display the notice "Powered by Hi3D 1.0". Violating laws, sanctions and export controls is prohibited.
Who it suits
The tool is aimed at those who work with generating 3D geometry from images and are ready to run the project from source in a CUDA environment. The open source code and published weights let you embed the generator into your own pipelines, and commercial use is permitted by the license — with the caveat of the scaling threshold and mandatory attribution.
This project is not for those without a CUDA-capable GPU or those looking for a ready-made installation package: the build is done from source, and there are no prebuilt wheel files. You should also factor in the size of the weights — 7.2 GB on disk — and the need to install flash-attn separately after PyTorch.
Hi3D v1.0 is a tool for specialists who need to generate 3D models from images with the option to refine the geometry. The project requires a Python 3.10 environment, a CUDA GPU and a willingness to build dependencies from source. The license permits commercial use but imposes obligations at scale and requires visible attribution in commercial products.
Fact sheet
Repository · Model on HuggingFace · Developer’s site
| Task | image-to-3d source |
|---|---|
| Code license | Hi3D Open License 1.0: commercial use allowed, separate license required at scale source |
| Weights license | Hi3D Open License 1.0: commercial use allowed, separate license required at scale source |
| Platform | CUDA per the author’s description source |
| Largest weights file | 3.4 GB source |
| All weights files | 7.2 GB source |
| Python | 3.10 per the author’s description source |
| Language | Python source |
| Last code change | 2026-10-23 source |
| Repository created | 2026-10-16 source |
| Latest release | v1.0 source |
|---|---|
| Release date | 2026-10-23 source |
| GitHub stars | 424 source |
| Forks | 32 source |
| Downloads per month | 0 source |
| Model updated | 2026-10-23 source |
Values are collected automatically from official sources and were checked on 2026-10-26. Each one links to its source, and values taken from the developer’s pages also carry a verbatim quote — hover over the note. Pricing and versions are shown as of the check date and change most often; verify on the vendor’s site before buying.



