WorldFoundry
English中文

VGGT-World

Method overview: Point cloud forecasting on TartanAir . Gen3R produces structurally disorganized geometry on walls and rooftops, whereas our method preserves coherent structure and yields more accurate predictions.

VGGT-World: world-model and 3d-scene-understanding; Runner parity pending.

Edit this pageGitHubMarkdown

VGGT-World builds a world model on top of VGGT-style 3D scene understanding: it couples geometry-grounded scene perception with video generation so rollouts stay consistent with the underlying 3D structure. WorldFoundry registers the official source route in-tree (paper: arXiv 2603.12655); checkpoints and assets are runtime-gated.

Official paper teaser: From video world models to geometry world models. Video world models predict future RGB in VAE latent space, coupling scene dynamics with appearance reconstruction. As a result, decoded predictions can remain.
Official paper teaser: From video world models to geometry world models. Video world models predict future RGB in VAE latent space, coupling scene dynamics with appearance reconstruction. As a result, decoded predictions can remain.
Method overview: Point cloud forecasting on TartanAir . Gen3R produces structurally disorganized geometry on walls and rooftops, whereas our method preserves coherent structure and yields more accurate predictions.
Method overview: Point cloud forecasting on TartanAir . Gen3R produces structurally disorganized geometry on walls and rooftops, whereas our method preserves coherent structure and yields more accurate predictions.

The recorded paper is VGGT-World: Transforming VGGT into an Autoregressive Geometry World Model (2026).

Code: SimonSun0810/VGGT-World.

The paper is VGGT-World: Transforming VGGT into an Autoregressive Geometry World Model. Code: SimonSun0810/VGGT-World.

Build your run

Pipeline
VGGTWorldPipelineworldfoundry.pipelines.world_model.pipeline_runtime_manifest
TaskDefault default
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  vggt-world \
  --pipeline.task-profile default \
  --json

What it is for

Use vggt-world for catalog world-model inference. Choose it when you need action-conditioned future frames rather than a standalone clip generator.

Beyond the recorded official repository (SimonSun0810/VGGT-World, pinned head c90f5ee2) and the arXiv paper pointer, architecture internals are not recorded in this repository.

Catalog tasks include world-model, 3d-scene-understanding, and video-generation. Documented outputs include video.

Recorded evaluation entry points include WorldScore (Unified world-generation quality across 3D-grounded rollouts).

Not VGGT and VGGT-Omega. Score on WorldScore.

Runtime contract

Artifacts generated_world and vggt_world_runtime_plan.json.

CLI task profiles: default. WorldFoundry binds VGGTWorldPipeline.

Family status: Runner parity pending.

Config resolution and pipeline import are expected to pass; real generation emits a runtime plan until required checkpoints/assets are staged. Missing assets are reported by the plan and are not treated as catalog blockers.

VGGT-World official source route is registered; checkpoints/assets are runtime-gated.

Running in WorldFoundry

The catalog launch is worldfoundry-eval run vggt-world --pipeline.task-profile default. 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).

No public checkpoint id is recorded on this card; follow the usage notes for staging. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh vggt-world. Add --download to fetch missing public checkpoints.

Status: Runner parity pending.