Flow - 4x faster than typing
Mini toolsLingBot World 2.0
By Robbyant
Model family: LingBot
LingBot-World 2.0, also called LingBot-World-Infinity, is the second iteration of Robbyant's LingBot-World interactive world model. A causal pretraining paradigm sustains an unbounded interaction horizon without quality loss over long rollouts. A distilled real-time variant enables 720p video generation at 60 fps for live interaction. The model supports a wide range of actions, including attacking, archery, spell-casting, and shooting, plus richer text-driven events than its predecessor. It introduces an agentic harness combining a pilot agent, which plans and executes character behaviors, and a director agent, which synthesizes new environmental elements as a scene unfolds. The released 14B causal-fast checkpoint takes an image and action sequence as input and outputs video, with inference code provided via generate.py.
New
Video
Overview
LingBot-World 2.0 is an interactive world model that generates real-time, controllable video worlds from a single image and action inputs. It supports diverse interactive actions (attacking, archery, spell-casting, shooting), an unbounded interaction horizon via causal pretraining, and a distilled real-time variant capable of driving 720p video at 60 fps. An agentic harness of pilot and director agents plans character behavior and synthesizes new environmental content as scenes progress.
Last updated: July 14, 2026
