WorldFoundry
English中文

Infinite-World

Method overview: Overview of Infinite-World architecture. (a) Hierarchical Pose-free Memory Compressor: The Hierarchical Pose-free Memory Compressor (HPMC) recursively compresses raw historical latents into a fixed memory budget via.

Infinite-World from Meituan (MeiGen-AI): world and video; Verified.

Edit this pageGitHubMarkdown

Infinite-World is MeiGen-AI's long-horizon world video model: from one image it navigates and extends a world over long durations. In this catalog it is verified — real GPU inference produced a non-empty long-horizon world video artifact with preview metadata through the Studio workspace job path, using the official image-conditioned navigation demo settings.

Method overview: Overview of Infinite-World architecture. (a) Hierarchical Pose-free Memory Compressor: The Hierarchical Pose-free Memory Compressor (HPMC) recursively compresses raw historical latents into a fixed memory budget via.
Method overview: Overview of Infinite-World architecture. (a) Hierarchical Pose-free Memory Compressor: The Hierarchical Pose-free Memory Compressor (HPMC) recursively compresses raw historical latents into a fixed memory budget via.

The recorded paper is arXiv 2602.02393.

Project page: rq-wu.github.io/projects/infinite-world. Code: MeiGen-AI/Infinite-World. Code: thuml/InfiniteWorld.

Build your run

Pipeline
InfiniteWorldPipelineworldfoundry.pipelines.infinite_world.pipeline_infinite_world
TaskInteractive Video interactive-video
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  infinite-world \
  --pipeline.task-profile interactive-video \
  --json

What it is for

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

The in-tree runtime resolves local or Hugging Face checkpoint layouts without relying on an external repository at inference time. Architecture internals beyond the recorded structure are not stored in this repository. It is published by Meituan (MeiGen-AI).

Catalog tasks include world, video, and world model. Documented outputs include video.

Recorded evaluation entry points include WorldModelBench (Long-horizon world-rollout quality), MemoBench (Long-horizon consistency evaluation).

Infinite-World long rollout. VGGT sibling is infinite-vggt.

Runtime contract

Optional prompt, negative_prompt, input_path, interactions, and frames; artifacts generated_world and infinite-world.mp4.

CLI task profiles: interactive-video. WorldFoundry binds InfiniteWorldPipeline.

Family status: Verified.

The profile matches the official local/Hugging Face checkpoint layout — start from the official image-conditioned demo settings that were verified.

In-tree Infinite-World runtime resolves local or Hugging Face checkpoint layouts without relying on an external repository at inference time.

Running in WorldFoundry

The catalog launch is worldfoundry-eval run infinite-world --pipeline.task-profile interactive-video. 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 thuml/InfiniteWorld. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh infinite-world. Add --download to fetch missing public checkpoints.

Status: Verified.