Skip to content
← Models
Open weights3DTopia/10 trustMetrics-checked

OpenLRM

Open-source reimplementation of the Large Reconstruction Model: single image to 3D NeRF in seconds.

Visit official page →

About

OpenLRM is a from-scratch open reimplementation of Adobe's LRM paper, built by Zexin He and Tengfei Wang with compute support from Shanghai AI Lab. It predicts a triplane NeRF from a single image in about five seconds and exports textured meshes or turntable videos. Small, base, and large variants were trained on Objaverse and MVImgNet, and the repo ships training code, Blender rendering scripts, and a Hugging Face demo.

The code is Apache-2.0, but the released weights carry a CC-BY-NC 4.0 license, so they are research-only and cannot be used commercially. That split license is the key gotcha: fine for study and prototyping, not for shipping products.

Specs

Context window

Single image input (224/336/448px by variant); mesh or video output

Pricing

Free for research (weights CC-BY-NC-4.0)

Free quota

Free — self-hosted; Hugging Face demo

API access

Self-hosted inference; Hugging Face demo

Weights

Open weights

Released

Not disclosed

Benchmarks

No published benchmarks; qualitative comparisons in the repo

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

image-to-3d3d-reconstruction3d-researchnerf

Trust score — 6/10

/10 trustMetrics-checked

Scored 6/10 (metrics tier, metrics-v1). Evidence: 1,256 GitHub stars with Apache-2.0 code confirmed via the repo page; reproduces the peer-reviewed LRM paper with training and inference code. Strengths: faithful open reimplementation with the full training pipeline published. Docked for: (1) weights are CC-BY-NC-4.0, research-only, blocking commercial use; (2) superseded by later feed-forward 3D models, with maintenance quiet since 2024. License: code Apache-2.0, weights CC-BY-NC-4.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.0/10
Maintenance · recency, releases, issue load
5.0/10
Manual review · sanity read + red-flag check
6.0/10
  • · 3DTopia (Zexin He, Tengfei Wang); Dec 2023; open reimplementation of Adobe LRM
  • · 1,256 stars; small/base/large variants trained on Objaverse + MVImgNet
  • · Single image to triplane NeRF in ~5s; training code + Blender scripts published
  • · Maintenance quiet since 2024; superseded by newer feed-forward models

Sources