Video Runtime
Developer-onlyFrame-StreamingOfficial source

LiveWorld

An interactive video world model that explicitly simulates dynamics outside the current camera view.

Access and release status are source-checked descriptions, not a live uptime check. A reachable website does not establish that video playback or interaction currently works.

Open Documentation ↗View GitHub ↗

No rights-cleared product preview yet. See the official page for its latest interface.

Published by Video Runtime · Source check 2026-09-14 · Evidence method · Suggest a correction

EDITORIALLY VERIFIED PRODUCT INFORMATION

About LiveWorld

An interactive video world model that explicitly simulates dynamics outside the current camera view. Combines an action-conditioned video backbone with state adapters so previously unseen or revisited regions can evolve beyond the visible viewport.

Key features

Releases training and inference code, processed examples and support for 1.3B and 14B configurations.

Getting started

Use Ubuntu 22.04, Python 3.11 and the documented CUDA/PyTorch stack, run the setup script to fetch weights and examples, then execute the included kid_coffee inference case.

Real-time interaction

Combines an action-conditioned video backbone with state adapters so previously unseen or revisited regions can evolve beyond the visible viewport.

Pricing

Apache-2.0 code; self-hosted compute costs extra Repository code is published under Apache-2.0. Confirm separate model, dataset and dependency terms before reuse. No hosted price is inferred. Budget for the documented GPU, storage and setup requirements.

Limitations

The tested setup lists H100 96 GB or H200 140 GB GPUs, at least 128 GB system memory and roughly 100 GB of pretrained assets.

Read the documentation ↗

EDITORIAL VIEW

Editor's Verdict

LiveWorld is included as a source-available implementation relevant to real-time generative video or interactive world systems.

Run the supplied example on representative hardware and review every linked licence before production use.

How It Works

  • Combines an action-conditioned video backbone with state adapters so previously unseen or revisited regions can evolve beyond the visible viewport.

What We Like

  • Releases training and inference code, processed examples and support for 1.3B and 14B configurations.

Current Limitations

  • The tested setup lists H100 96 GB or H200 140 GB GPUs, at least 128 GB system memory and roughly 100 GB of pretrained assets.

Pricing & Access

  • Repository code is published under Apache-2.0. Confirm separate model, dataset and dependency terms before reuse.
  • No hosted price is inferred. Budget for the documented GPU, storage and setup requirements.

Verification Summary

  • Official repository and licence reviewed on 2026-09-14. Video Runtime has not installed or benchmarked this project.
  • Performance and compatibility statements below are attributed to the maintainers and are not site measurements.
Technical notes & official performance

Official Performance Data

  • The repository frames LiveWorld as an interactive world model but does not publish one comparable end-to-end latency figure for the released setup.

Technical Notes

  • Flash Attention is required; training and sample creation support multi-node multi-GPU execution.

Sources

  1. Official GitHub repository ↗github · accessed 2026-09-14
  2. Code licence ↗github · accessed 2026-09-14
  3. Official project page ↗official-docs · accessed 2026-09-14

Access, stage & mechanism

These are separate source-checked facts, not a live uptime monitor. Unverified fields are left open; transport and persistent sessions do not establish how frames are generated.

Release stage
Not verified
Access mode
Not verified
Source availability
Not verified
Transport
Not verified
Interaction
Not verified
Session mode
Not verified
Generation mechanism
Not verified
Continuity method
Not verified
Licence status
Not verified
Tasks
Not classified

Read how official-source and hands-on records stay separate in our Methodology.