@FinanceYF5: 1/ Most AI "world models" suffer from visual collapse and scene drift after just a few minutes of running. @robbyant_brain just open-sourced LingBot-World 2.0 (Infinity): supports hour-level real-time generation, no quality degradation over long runs, stable 720p/60fps output. This is...

1/ Most AI “world models” start glitching and drifting scenes after just a few minutes.

@robbyant_brain just open-sourced LingBot-World 2.0 (Infinity): supports hour-long real-time generation, with no quality degradation during extended runs, stably outputting 720p/60fps.

This is a key step in the “see-act-predict-think” full-stack chain, where robots learn to “anticipate the world.” 👇 https://t.co/g9sWAjUCGu

2/ First, the hardest part: long-term generation without “breaking”

Most interactive world models end up with blurry textures and drifting scenes after just a few minutes. LingBot-World 2.0 supports infinite scene extension, with stable 720p/60fps output and no quality loss or scene collapse during continuous long runs.

3/ Interaction actions are also upgraded

Compared to the previous generation, this version adds broader action and event types, unlocking more diverse interaction mechanics — no longer just walking around and observing, but truly participating in and changing the scene.

4/ The world “grows on its own”

The built-in Director Agent proactively suggests new events to users in real time — a blizzard, a flock of birds suddenly appearing — the world is no longer a static background, but a continuously evolving system.

It now also supports multiple people entering the same world together: one person controls the exploration, while another can play the “director” role, guiding events in real time.

5/ 14B model, officially open-sourced today

LingBot-World 2.0 weights are now available. Developers can deploy it on their own using SGLang (day-0 support), or try it directly without installation on Reactor or LingGuang App. GitHub: http://github.com/Robbyant/lingbot-world-v2… Weights: http://huggingface.co/collections/robbyant/lingbot-world-v2… Official site: https://technology.robbyant.com/lingbot-world-v2… Online demo: http://reactor.inc/lingbot-world-v2…

6/ This is not the first time @robbyant_brain has done this.

Just like in January, this is another open-source release — three models have been released so far: spatial perception (LingBot-Vision + LingBot-Depth 2.0), embodied manipulation (LingBot-VLA 2.0), and today’s world simulator LingBot-World 2.0.

A technical pipeline from “seeing the world” to “predicting the world” is being gradually laid out. It’s worth watching what comes next.

That’s all for now.

If you like this topic:

1. Follow me (@FinanceYF5)
2. Like + retweet the first post below