Hi3D 1.0

Hi3D 1.0 (formerly Sparc3D) is an image-conditioned 3D geometry refinement model open-sourced by the Hi3D team in Beijing on October 23, 2026.

It is pitched as production-validated, turning coarse meshes into high-fidelity geometry.

What it is

• Role: refines/reconstructs 3D asset geometry conditioned on a reference image
• Training data: ~2M curated 3D assets
• Resolution: up to ~1536³ effective resolution, preserving fine structures
• Open source: code at GitHub hi3d-official/Hi3Dv1.0, weights on Hugging Face

Why it matters

High-resolution 3D generation is bottlenecked by memory and data. Hi3D opens up the sparse-representation approach of Sparc3D, and spread on X as "Hi3D made its full AI 3D model FREE".

How to use it

1. Try it free online at hi3d.ai/hi3d-1-0-open-source
2. Run locally: clone github.com/hi3d-official/Hi3Dv1.0 and download weights from Hugging Face (a high-VRAM GPU is needed)
3. Typical flow: reference image + coarse mesh → refined geometry, then texture in Blender etc.

Good for: game/film assets, 3D printing, e-commerce 3D views.

Check the repo LICENSE before commercial use.