According to Latent Space, Runway's GWM Worlds 2 introduces WorldPrompt, a method for steering a world model that generates video and audio in real time. The key idea is that the model maintains persistent context, meaning it remembers what has already been generated and uses that memory to inform subsequent frames. This is a departure from typical text-to-video tools that produce a fixed clip from a single prompt.
The system also supports timed actions, allowing creators to specify events or changes at particular moments rather than only describing a static scene. This makes it possible to direct a continuous, evolving world rather than just a one-shot output. The article frames this as an engineering step toward more interactive and controllable generative environments.
Because this is based on a single source, the details reflect Latent Space's reporting on Runway's announcement. No independent verification or contrasting perspectives are included here, but the described mechanism highlights a growing focus on real-time, persistent world models in AI media generation.