The competition to develop world models is intensifying, with AI image and video generation firm Runway now entering the arena alongside a growing list of startups and major technology corporations. The company has introduced its inaugural model, named GWM-1. According to Runway, this model operates by predicting frames sequentially, crafting a simulation that comprehends physics and the genuine behavior of the world over time.

A world model is an AI system designed to learn an internal simulation of how the world functions. This capability allows it to reason, plan, and act without requiring training on every conceivable real-life scenario. Runway, which recently launched its Gen 4.5 video model that outperformed both Google and OpenAI on the Video Arena leaderboard, stated that its GWM-1 world model is more "general" than competitors like Google's Genie-3. The company positions it as a tool for creating simulations to train agents across various fields, including robotics and life sciences.

"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," explained the company's CTO, Anastasis Germanidis, during a livestream event.

Runway has released specific specialized versions of the new world model, called GWM-Worlds, GWM-Robotics, and GWM-Avatars.

GWM-Worlds is an application that enables users to create interactive projects. By setting a scene through a text prompt or an image reference, the model generates a world with an understanding of geometry, physics, and lighting as the user explores the space. The company noted that these simulations run at 24 frames per second and 720p resolution. While GWM-Worlds has potential applications in gaming, Runway also highlighted its suitability for teaching agents how to navigate and behave in the physical world.

With GWM-Robotics, the aim is to utilize synthetic data enhanced with new parameters, such as varying weather conditions or obstacles. Runway suggests this approach could also help identify when and how robots might violate policies and instructions in different scenarios. Additionally, the company is developing realistic avatars under GWM-Avatars to simulate human behavior. This area has seen work from other companies like D-ID, Synthesia, Soul Machines, and Google, focusing on creating lifelike avatars for communication and training purposes.

Runway clarified that, technically, Worlds, Robotics, and Avatars are separate models, but the long-term plan is to merge them into a single unified model.

Beyond the new world model, Runway is also updating its foundational Gen 4.5 model, which was released earlier in the month. This update introduces native audio and long-form, multi-shot generation capabilities. Users can now generate one-minute videos featuring character consistency, native dialogue, background audio, and complex shots from multiple angles. The model also allows for editing existing audio, adding dialogues, and modifying multi-shot videos of any length.

The Gen 4.5 update brings Runway closer to the capabilities of competitor Kling's all-in-one video suite, which also launched recently, particularly in the areas of native audio and multi-shot storytelling. This advancement signals a shift for video generation models from prototype stages to production-ready tools. Runway's updated Gen 4.5 model is accessible to all users on paid plans.

The company announced that GWM-Robotics will be made available through a Software Development Kit (SDK). Runway added that it is actively engaged in discussions with several robotics firms and enterprises regarding the use of both GWM-Robotics and GWM-Avatars.