Runway publishes GWM Worlds 2, a real-time interactive world model

Background

Generative video systems often begin with a prompt and return a finished clip. A world model takes a different approach: it maintains a simulated scene that can be explored and changed while it is running. Runway published GWM Worlds 2 on 3 September 2026 as an example of that approach.^[1] The company describes it as a research preview that extends GWM Worlds with generated audio and richer subject and scene control.^[1]

What the model does

GWM Worlds 2 turns high-fidelity video and audio generation into real-time interactive simulation.^[1] A user first defines a world by specifying its environment, subjects, visual style, physical rules and ambience.^[1] Those choices establish what the world contains, how it looks, and how it is meant to behave. After the world is defined, the user steers it through free-form text actions and continuous camera motion.^[1] The text supplies instructions in natural language, while camera motion provides a spatial path through the scene. Together, they make interaction part of generation rather than a one-time prompt.

Output and interaction

While the user explores, the model generates continuous 720p video at 24 frames per second and audio at 48,000 Hz.^[1] The video and audio respond to the user's inputs as the world is explored.^[1] In practical terms, 720p describes the resolution of the image, 24 frames per second describes how many video frames are produced each second, and 48,000 Hz describes the audio sampling rate. The combination of visual and audio output is central to the interactive simulation: the world is not only seen but heard, and both streams follow the user's actions.

Why it matters

Runway presents GWM Worlds 2 as a research preview, so it is a step in an ongoing line of work rather than a declared finished product.^[1] Its stated advance over GWM Worlds is generated audio plus richer control over subjects and scenes.^[1] The broader importance is in the type of system it demonstrates: video and audio generation used as an interactive simulation that a user can direct in real time.^[1] The user does not only receive a generated result; they define a world and then steer it with text and camera movement.^[1] That shift places control at the center of the experience, showing how generation can be organized around an explorable world rather than a fixed output.