Runway is expanding from AI video generation into world models, a category of AI systems designed to simulate how environments behave over time. Its first model in this area, GWM-1, is being presented as a step toward general-purpose simulation for interactive worlds, robotics and avatars.
The launch comes alongside an update to Runway’s Gen 4.5 video model. That update adds native audio and longer, more complex video generation, moving the company closer to tools that can support complete video workflows rather than short demonstrations.
What Runway means by a world model
A world model is an AI system that learns an internal simulation of how the world works. The goal is to let the system reason, plan and act without requiring training on every possible real-world situation.
Runway says GWM-1 works through frame-by-frame prediction. In practical terms, that means the model creates a simulation by predicting what should happen next across time, with an understanding of physics and how the world behaves.
The company’s argument is that strong video generation is the foundation for broader simulation. During the livestream, CTO Anastasis Germanidis said, “To build a world model, we first needed to build a really great video model. We believe that the right path to building a world model is teaching models to predict pixels directly is the best way to achieve general-purpose simulation. At sufficient scale and with the right data, you can build a model that has sufficient understanding of how the world works.”
That framing matters because Runway is not treating GWM-1 only as a media tool. It is pitching the model as a way to create simulated environments that could be used to train agents in areas such as robotics and life sciences.
Three versions for worlds, robots and avatars
Runway released specific slants or versions of GWM-1: GWM-Worlds, GWM-Robotics and GWM-Avatars. The company noted that these are technically separate models for now, while also saying it plans to merge them into one model eventually.
GWM-Worlds is an app that lets users create an interactive project. A user can define a scene through a prompt or an image reference, then explore the space as the model generates the world around them.
Runway says the model generates that world with an understanding of geometry, physics and lighting. The company also said the simulation runs at 24 fps and 720p resolution.
The most obvious use case for GWM-Worlds is gaming, and Runway acknowledged that possibility. But the company is also positioning it as a way to teach agents how to navigate and behave in the physical world.
Why robotics is a major focus
GWM-Robotics is aimed at synthetic data. Runway says it can enrich that data with new parameters, including changing weather conditions or obstacles.
That is important because robotics systems need to handle variation. A simulated setting can expose a system to situations that may be hard, costly or impractical to recreate repeatedly in the physical world.
Runway also says this approach could reveal when and how robots might violate policies and instructions in different scenarios. That makes GWM-Robotics not only a training tool, but also a way to examine behavior before deployment.
The company said GWM-Robotics will be made available through an SDK. It also said it is in active conversation with several robotics firms and enterprises about using GWM-Robotics and GWM-Avatars.
Avatars and human behavior simulation
GWM-Avatars is Runway’s effort to build realistic avatars that simulate human behavior. The source places this work in a broader field that includes companies such as D-ID, Synthesia, Soul Machines and Google, all of which have worked on human avatars for uses including communication and training.
For Runway, avatars fit the same larger idea as interactive worlds and robotics: simulated behavior that can be generated, tested and adapted. In this case, the focus is not the movement of a robot or the physics of a generated environment, but the behavior of realistic human representations.
The company has not described these three tracks as permanently separate. Its plan to merge Worlds, Robotics and Avatars into one model suggests a broader ambition: a single simulation system that can cover environments, machines and human-like characters.
Gen 4.5 gets native audio and longer videos
Runway is also updating Gen 4.5, the video model it launched earlier in the month. The source says Gen 4.5 surpassed both Google and OpenAI on the Video Arena leaderboard.
The new Gen 4.5 update adds native audio and long-form, multi-shot generation. Runway says users can generate one-minute videos with character consistency, native dialogue, background audio and complex shots from various angles.
The update also allows users to edit existing audio and add dialogue. In addition, Runway says users can edit multi-shot videos of any length.
These features push Gen 4.5 toward more complete video production. Instead of generating isolated clips, the model is being extended toward scenes with continuity, audio, dialogue and multiple camera angles.
The source compares this move with competitor Kling’s all-in-one video suite, which also launched earlier this month, especially around native audio and multi-shot storytelling. It also reads as another sign that video generation models are moving from prototype tools toward production-ready systems.
Runway’s updated Gen 4.5 model is available to all paid plan users.