The landscape of video generation is undergoing a significant transformation, moving beyond simply creating visually appealing clips to building virtual environments that support interaction and maintain physical plausibility. This crucial development points toward the emergence of video foundation models that function implicitly as world models. These world models, which aim to simulate the real world, are sophisticated digital engines that encode comprehensive world knowledge to simulate real-world dynamics in accordance with intrinsic physical and mathematical laws. A modern video foundation model is conceptualized as the combination of two core components: an implicit world model and a video renderer. The world model serves as a latent simulation engine, encoding structured knowledge about physical laws, interaction dynamics, and agent behavior, enabling coherent reasoning and goal-driven planning. The video renderer then translates this latent simulation into realistic visual observations, providing a “window” into the simulated world. The foundation of this shift lies in how humans and embodied agents perceive reality: vision is the dominant sensory modality through which we learn and reason about the world. This intrinsic reliance on visual representation makes video generation an information-rich foundation for constructing world models. The evolution of this sophisticated use of Artificial Intelligence can be traced through four generations, advancing capabilities such as faithfulness, interactiveness, and complex task planning. Current research shows progress toward models (Generation 3 and 4) achieving physically intrinsic faithfulness and complex task planning, capable of simulating complex systems like weather patterns or narrative plots. These systems act as high-fidelity simulators for domains such as robotics, autonomous driving, and interactive gaming. Ultimately, world models driven by AI promise to support high-stakes decision-making and advance autonomous systems by creating virtual environments that simulate everything, everywhere, and anytime.