LingBot-World 2.0: Interactive world generation from image, text, and actions (github.com)

LingBot-World 2.0, also referred to as LingBot-World-Infinity, has been unveiled with significant advancements including an unbounded interaction horizon and rapid response times, enabling real-time interaction capable of driving 720p video streams at 60 fps. Notably, the update expands the diversity of interactive elements, introducing a wide array of actions such as attacking and spell-casting alongside a richer set of text-driven events. The introduction of an agentic harness marks a critical innovation, allowing a planning agent to manage character behavior and a director agent to generate environmental elements dynamically. These improvements are pivotal for the AI and machine learning communities as they enhance the complexity and interactivity of AI-driven simulations, facilitating more immersive experiences. The release includes a variety of models and inference codes, promoting accessibility and integration for researchers interested in advanced world modeling techniques. This project not only demonstrates the capabilities of causal pretraining but also emphasizes the collaborative nature of AI development, thanks to its foundational ties with the open-source contributions from the Wan Team. The official demo will be showcased at WAIC 2026, inviting further exploration of its vast potential.