---
title: "VGGT-Omega"
description: "VGGT-Omega from Meta AI: 3d-reconstruction and gaussian-splatting; Verified."
modelId: "vggt-omega"
pageSource: authored
---

VGGT-Omega is the extended VGGT release published under facebook/VGGT-Omega, covering 3D reconstruction, Gaussian splatting, and novel view synthesis. In this catalog it is verified with unusually strong evidence: single-image GPU parity against the pinned upstream commit passed with max_abs_diff=0.0 for comparable outputs, and the in-tree runner supports the official scene-export route.

<ModelPaperFigures modelId="vggt-omega" locale="en" />

The recorded paper is [arXiv 2605.15195](https://arxiv.org/abs/2605.15195).

Project page: [vggt-omega.github.io](https://vggt-omega.github.io/). Code: [facebookresearch/vggt-omega](https://github.com/facebookresearch/vggt-omega).

<ModelCommandBuilder modelId="vggt-omega" locale="en" />

## What it is for

`vggt-omega` is the recorded reconstruction line. Start here to emit geometry from the documented views.

It uses an official-architecture representation runner plus the existing WorldFoundry VGGTResult/save interface, emitting predictions.npz plus GLB scene artifacts. Weights stay under assets/checkpoints and are never committed into source files; the checkpoint license is CC-BY-NC-4.0. It is published by Meta AI.

Catalog tasks include 3d-reconstruction, gaussian-splatting, and novel-view-synthesis. Catalog aliases include `vggt_omega` and `facebook/VGGT-Omega`.

Not [VGGT](/docs/guides/supported-models/vggt) and [VGGT-World](/docs/guides/supported-models/vggt-world).

## Runtime contract

Optional `input_path`, `interactions`, `camera_view`, `task_type`, and `image_resolution`; artifacts `scene_conf20.0_blackFalse_whiteFalse_camTrue_skyFalse_max200k.glb`.

CLI task profiles: `interactive-video`. WorldFoundry binds `VGGTOmegaPipeline`.

Family status: **Verified**.

Use this entry for the verified official scene-export route with WorldFoundry representation artifacts. It is distinct from the vggt (base) and vggt-world entries — pick by route, not by name.

VGGT-Omega uses an official-architecture representation runner plus the existing WorldFoundry VGGTResult/save interface.

## Running in WorldFoundry

Launch `worldfoundry-eval run vggt-omega --pipeline.task-profile interactive-video` to write geometry. The command builder above switches `--pipeline.task-profile` among recorded tasks and still emits `worldfoundry-eval run`.

It runs in the unified environment `worldfoundry-unified-cu128` (Python 3.11, CUDA 12.8).

Recorded weight repositories include `facebook/VGGT-Omega`. Prepare the environment and check local assets with `bash scripts/inference/prepare_model_infer.sh vggt-omega`. Add `--download` to fetch missing public checkpoints.

Recorded licenses include cc-by-nc-4.0; status: **Verified**.

<ModelRelatedRecipes modelId="vggt-omega" locale="en" />