AnswersPublished20 sources
Hi3D v1.0 Is a 3D Mesh Refiner, Not a Generator From Scratch
Hi3D v1.0 was released on October 23, 2026—not October 28—with inference code on GitHub, weights on Hugging Face and an online demo.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Hi3D v1.0 was released on October 23, 2026—not October 28—with inference code on GitHub, weights on Hugging Face and an online demo.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Chinese 3D generation company Hi3D open-source in its October 28, 2026 release of Hi3D v1.0, including the availability of inferenc. Article summary: Hi3D released Hi3D v1.0 as a developer-facing, image-conditioned 3D geometry-refinement package, with inference code on GitHub, model weights on Hugging Face and an online demo—not simply an object generator that starts . Topic tags: general, general web. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clic
Hi3D v1.0 is an image-conditioned 3D geometry refiner: it takes a coarse mesh and a reference image of the same object, then progressively builds more detailed geometry. The release included inference code, model artifacts and documentation, with a demo available online. The official GitHub release is dated October 23, 2026; October 28 was the date of later coverage. 8
11
13
44
What Hi3D released
Developers can find the inference code and release details on GitHub, download model weights and configurations from Hugging Face, or try an online demo. Hi3D’s announcement described the demo as free to use. 8
11
13
44
The practical distinction is that Hi3D v1.0 handles the refinement stage, not the entire image-to-3D process. It expects a coarse mesh as an input alongside an image of the same object; creating that initial mesh is a separate step. 8
18
How the refinement pipeline works
Hi3D v1.0 retains the Sparc3D representation and refinement architecture. In each pass, the pipeline samples latent representations at occupied voxel locations, decodes them into geometry with a self-pruning sparse VAE, and extracts a mesh. That mesh can then provide sparse support for another refinement pass. 3
18
The release describes three refinement levels:
| Passes | Nominal sparse extraction-grid resolution |
|---|---|
| 1 | 512³ |
| 2 | 1024³ |
| 3 | 1536³ |
These are grid-resolution figures, not fixed output triangle counts. The project documentation describes the third pass as extending refinement to a nominal 1536³ sparse extraction grid. 8
18
The project README reports training on approximately two million curated 3D assets. That is a figure reported by the project, rather than an independently verified dataset audit. 3
Hardware and validation caveats
The installation documentation specifies Python 3.10, an NVIDIA GPU with compute capability 8.0 or newer, and a compatible CUDA toolkit version in the 12.0–12.5 range. 17
The same documentation says the 1536³ refinement ladder has been validated on a 96 GB NVIDIA H20 and that peak memory use and feasibility on smaller cards have not been established. That makes 96 GB a reported validation setup—not a published minimum requirement for every resolution. 17
Read the license before commercial use
Hi3D’s materials are governed by the Hi3D Open License Agreement Version 1.0, which grants specified rights to use, reproduce, modify, fine-tune and adapt covered materials subject to the agreement’s conditions. The license does not replace the original terms and attributions that apply to third-party components. 6
8
The license materials describe a scale threshold that can be triggered by more than 200,000 monthly active users or monthly revenue of at least US$1 million. A reproduced copy of the agreement states that users must notify the licensor and apply for a separate license within 60 days after triggering that threshold. Check the current agreement and its scope before integrating the model into a commercial product. 30
43
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Hi3D v1.0 Is a 3D Mesh Refiner, Not a Generator From Scratch"?
Hi3D v1.0 was released on October 23, 2026—not October 28—with inference code on GitHub, weights on Hugging Face and an online demo.
Sources
github.comReleases: hi3d-official/Hi3Dv1.0 - GitHub
huggingface.cohi3d-official/Hi3Dv1.0 at main - Hugging Face
prnasia.comHi3D V1.0 Goes Open Source: The First Production-Validated Open-Source AI 3D Model