Robbyant · 2026-07-09 · notable
LingBot-World 2.0 — Robbyant open-sources a 14B interactive world model
LingBot-World 2.0 is a real-time interactive world model that renders 720p video at 60 fps and responds to actions like attacking and spellcasting. Robbyant released the paper, code, and a 14B causal-fast checkpoint on July 9 under CC BY-NC-SA 4.0.
Robbyant's LingBot-World 2.0 turns a text prompt into an infinite, playable video world at 720p and 60 fps.
What is it?
LingBot-World 2.0 is Robbyant's second-generation interactive world model, released on July 9 as arXiv paper 2607.07534 with code and a 14B causal-fast checkpoint under CC BY-NC-SA 4.0. The system generates video that a user or agent can act inside — attacking, shooting an arrow, casting a spell — while keeping the scene consistent across an unbounded horizon.
How does it work?
A causal pretraining recipe lets the primary 14B model keep interaction consistent frame-to-frame, and a distilled real-time variant drives 720p output at 60 fps. Robbyant pairs the video model with planning and environment-synthesis agents that expand the world as the player moves, and adds multiplayer support so multiple people can share the same generated scene.
Why does it matter?
Interactive world models are the missing piece between video generation and playable AI experiences. LingBot-World 2.0 is the first open-weights model to hit 720p 60 fps with a broad action vocabulary, so game and simulation teams can prototype without a closed API — the licence is non-commercial, but research and demo use are free.
Who is it for?
Game researchers, robotics simulation teams, and generative-video labs studying interactive scene models.
Try it
git clone https://github.com/Robbyant/lingbot-world-v2 and pull huggingface.co/robbyant/lingbot-world-v2-14b-causal-fast