README.md

metadata

title: Hi3Dv1
emoji: 🗿
colorFrom: gray
colorTo: purple
sdk: gradio
sdk_version: 6.30.0
app_file: app.py
python_version: '3.10'
short_description: Image to high-res 3D mesh with Hi3Dv1 refinement
startup_duration_timeout: 1h
license: other
license_name: hi3d-open-license-agreement-1.0
license_link: LICENSE
models:
  - hi3d-official/Hi3Dv1.0
  - tencent/Hunyuan3D-2mini
  - facebook/dinov2-large
  - openai/clip-vit-large-patch14

Hi3Dv1 demo

Image → coarse mesh (Hunyuan3D-2mini-Turbo, or upload your own) → high-resolution geometry refinement with hi3d-official/Hi3Dv1.0.

Powered by Hi3D 1.0 — http://hi3d.ai/

• The sparc3d/ package is the official inference code from https://github.com/hi3d-official/Hi3Dv1.0 (Hi3D Open License Agreement 1.0, plus retained Apache-2.0/MIT terms — see LICENSE, NOTICE, THIRD_PARTY_LICENSES.md, AUTHORS). One patch: sparse convolutions use spconv's Native algorithm (SPCONV_ALGO=native) so they run on ZeroGPU's Blackwell GPUs.
• Advanced settings → "Also show the coarse (intermediate) mesh" reveals the coarse mesh the pipeline refines (viewer + GLB download) next to the final result; it is hidden by default.
• Hunyuan3D-2mini is a third-party model under the Tencent Hunyuan Community License.
• Example images come from microsoft/TRELLIS (MIT License, © Microsoft Corporation).