WorldFoundry
English中文

InSpatio-World

Method overview: InSpatio-World : Toward a Versatile 4D World Simulator. Top: Our framework enables the synthesis of diverse dynamic scenes from a single video, supporting real-time, high-DoF interactive 4D roaming experiences..

InSpatio-World: video-conditioned-world-generation and trajectory-conditioned-video; Verified.

Edit this pageGitHubMarkdown

InSpatio-World is a 1.3B video-conditioned world generator: given a video and a target trajectory it synthesizes trajectory-conditioned novel views of the same world. In this catalog it is fully verified — the in-tree pipeline generated and visually validated a 297-frame video through the Workspace on 2026-07-10, with every default component resolving from local checkpoint storage.

Method overview: InSpatio-World : Toward a Versatile 4D World Simulator. Top: Our framework enables the synthesis of diverse dynamic scenes from a single video, supporting real-time, high-DoF interactive 4D roaming experiences..
Method overview: InSpatio-World : Toward a Versatile 4D World Simulator. Top: Our framework enables the synthesis of diverse dynamic scenes from a single video, supporting real-time, high-DoF interactive 4D roaming experiences..

The recorded paper is arXiv 2604.07209.

Project page: inspatio.github.io/inspatio-world. Code: inspatio/inspatio-world. Code: madebyollin/taehv.

Build your run

Pipeline
InspatioWorldPipelineworldfoundry.pipelines.inspatio_world.pipeline_inspatio_world
TaskDefault default
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  inspatio-world \
  --pipeline.task-profile default \
  --pipeline.input-path VALUE \
  --json

What it is for

inspatio-world is the recorded reconstruction line. Start here to emit geometry from the documented views.

The recorded checkpoint is inspatio/world (apache-2.0); official inference source is retained in-tree. An optional use_tae path resolves taew2_1.pth from the shared checkpoint root but is not required for the default Wan VAE path.

Catalog tasks include video-conditioned-world-generation, trajectory-conditioned-video, and novel-view-synthesis. Catalog aliases include inspatio-world-1.3b.

Recorded evaluation entry points include CameraBench (Trajectory-following accuracy of generated views).

Runtime contract

Required input_path; optional prompt, traj_txt_path, skip_step1, skip_step2, and skip_step3; artifacts inspatio-world.mp4.

CLI task profiles: default. WorldFoundry binds InspatioWorldPipeline.

Family status: Verified.

Every default component resolves from local checkpoint storage only — stage the inspatio/world weights first, then run the standard pipeline.

The in-tree InSpatio-World pipeline generated and visually validated a 297-frame video through Workspace on 2026-07-10.

Running in WorldFoundry

Launch worldfoundry-eval run inspatio-world --pipeline.task-profile default 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 inspatio/world, Wan-AI/Wan2.1-T2V-1.3B, and depth-anything/DA3NESTED-GIANT-LARGE-1.1 among others. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh inspatio-world. Add --download to fetch missing public checkpoints.

Recorded licenses include apache-2.0; status: Verified.