Hi3D open-sources V1.0 image-to-3D model and launches Hi3D Agent
Overview
Hi3D has open-sourced its V1.0 image-to-3D model on GitHub and Hugging Face and launched Hi3D Agent, a creator workflow.
V1.0 refines a rough mesh into detailed geometry using a reference image, and outputs shape only, with no colors or textures. Elvis Saravia says texture generation is available only in the commercial Hi3D V3.
The model requires Linux and an NVIDIA GPU with compute capability 8.0 or newer, according to Pandaily, which reports that training code is not included. Pandaily and Rohan Paul's post describe a commercial license that requires notifying Hi3D and obtaining a separate license for products above 200,000 monthly active users. Rohan Paul's post also cites a $1M monthly revenue threshold.
Hi3D reports an 85.9% preference score among professional 3D practitioners and a 90% slicing pass rate at 1536³ resolution, against 7% to 17% for competitors. Rohan Paul's post reports V1.0 scored 0.75 on multi-view fidelity and 0.841 on detail richness against eight other models. These figures come from Hi3D and have not been independently verified.
Written by AI from the articles below · updated Oct 28, 3:22 AM ET
Check the sources:
Developments
2 developments
- Oct 27, 4:18 AM ET · 4 articlesHi3D open-sources Hi3D v1.0 image-to-3D modelQbitAI: Hi3D open-sources its V1.0 3D generation model with a 2M-asset training set
- Oct 27, 3:59 AM ET · 1 articleHi3D open-sources its V1.0 production-grade 3D generation base model and launches Hi3D AgentGeekPark: Hi3D open-sources its V1.0 3D model and launches a creator Agent
Article timeline
The articles in this story. Times are ET.
- PandailyNewsHi3D open-sources Hi3D v1.0, a Sparc3D-based model that refines 3D meshes to 1536 resolution
AIChinese 3D generation company Hi3D has open-sourced Hi3D v1.0, which refines a coarse mesh using a reference image of the same object. The model is built on Sparc3D, expanded to about two million curated 3D assets, and requires Linux and an NVIDIA GPU with compute capability 8.0 or newer. Training code is not included, and the release allows commercial use under its own license with a notice requirement above 200,000 monthly active users.
elvis@omarsar0XHi3D releases open-source Hi3D V1.0 for 3D mesh refinementAIHi3D released Hi3D V1.0, an open-source 3D model built on Sparc3D that adds detail to a rough 3D mesh using a reference image. V1.0 includes inference code and model weights for geometry refinement, while texture generation is available only in the commercial Hi3D V3.

Rohan Paul@rohanpaul_aiXHi3D v1.0 open-sources an image-to-3D model that holds shape from every angleAIHi3D has open-sourced Hi3D v1.0, an image-to-3D model that refines a coarse mesh in three passes at 512, 1024 and 1536 resolution. The post reports it scored best on multi-view fidelity (0.75) and detail richness (0.841) against eight other models, and was picked 83.5% of the time in 562 side-by-side ratings. Code and weights are on GitHub and Hugging Face, free for commercial use up to 200K monthly users or $1M monthly revenue.

- QbitAINewsHi3D open-sources its V1.0 3D generation model with a 2M-asset training set
AIHi3D has open-sourced its Hi3D V1.0 3D generation model on GitHub and Hugging Face, built on the Sparc3D architecture. Its training data was expanded to about 2 million curated 3D assets, four times the 500,000 in the Sparc3D paper. The company reports a 90% slicing success rate and first-place results on multi-view fidelity and detail richness among the methods it compared.
- GeekParkNewsPickHi3D open-sources its V1.0 3D model and launches a creator Agent
AIHi3D has open-sourced its production-validated V1.0 3D generation model and launched Hi3D Agent, a creator-focused workflow. The company says V1.0 reached an 85.9% preference score among professional 3D practitioners and a 90% slicing pass rate at 1536³ resolution, versus 7% to 17% for competitors. Hi3D has raised nearly $50 million in two rounds within six months.
Heat Trend
- Current comparable heat
- 30
- Peak comparable heat
- 34Oct 27, 1:00 PM ET
- Change in past 24 hours
- +58%
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.