New Delhi: Runway has officially announced the release of its first proprietary global model named GWM-1. This system will operate on the principle of step-by-step, frame-by-frame forecasting, by creating a simulation with an understanding of the physics and how the world evolves over time. A world model is an AI that learns internal models of how the world works to reason, plan, and act without needing training on each individual real-world scenario. Runway has released the Gen 4.5 video model, which outpaced competitors in the Video Arena standings, including Google and OpenAI.
The company also claims that the GWM-1 is more versatile than the Genie-3 and other competitors. It is positioned as a tool for creating simulations that can train agents across the range of domains, from robotics to biosciences. Runway also unveiled separate branches for the latest world model under the names GWM-Worlds, GWM-Robotics and GWM-Avatars. GWM-Worlds is an extension of the model that allows an interactive project; the user defines a scene with the prompt, and as they explore, the model constructs a world with an understanding of geometry, physics, and lighting.
The company notes that the Worlds can be helpful for both gaming and training agents in navigation and behaviour within a physical environment. GWM-Robotics focuses on using synthetic data with the latest parameters, such as the changing weather conditions or the presence of obstacles. Runway stated this approach can also detect when and how robots might violate policies or instructions in different scenarios.
GWM-Avatars are being developed to model realistic human behaviour. In the lineup of the cutting-edge giants in the avatar space, names such as D-ID, Synthesia, and Soul Machines, as well as Google, have previously worked, but Runway emphasises its own approach to creating convincing human avatars for communication and training.
