Runway Introduces GWM Worlds 2 for Real-Time Interactive World Generation
Runway has unveiled research on GWM Worlds 2, a General World Model capable of generating real-time interactive video and audio simulations at continuous 720p at 24 fps with 48 kHz audio. The system introduces WorldPrompt, which combines a persistent genesis prompt with continuous camera inputs and timestamped text actions to steer the generation live without preset clip lengths. This shift from producing static video clips to running steerable, real-time world models represents a major leap forward for interactive entertainment, gaming, and virtual production. Furthermore, the ability to interactively navigate persistent synthetic environments provides critical foundation tools for embodied AI and robotics simulation. Technically, GWM Worlds 2 relies on an autoregressive diffusion architecture utilizing sliding KV-cache attention alongside causal video and audio decoders. Frame generation is conditioned on global context (the genesis prompt and initial frame), current inputs (active text actions and camera controls), and a sliding window of cached past frames.
## BACKGROUND
World models in artificial intelligence are generative models trained on vast visual datasets to simulate physical environments, predicting how scenes unfold based on user or agent actions. Traditional generative video models generate fixed-duration clips from static text prompts without allowing continuous user interaction or state persistence.