Runway

Runway's GWM Worlds 2 is a General World Model that turns audio and video generation into an interactive simulation a user can play in real time, aimed at researchers, filmmakers, game designers, and robotics teams who need controllable synthetic environments. The model generates 720p video at 24 frames per second and 48 kHz audio, one step at a time , and it follows a "world prompt" split into a persistent scene with its subjects and rules plus a time-stamped event stream of text actions, with per-frame camera motion as a separate continuous input. Under the hood, Runway fine-tunes its foundational audio-visual model on this world prompt format and then post-trains the bi-directional version into an auto-regressive one with no fixed duration, so sessions can continue indefinitely rather than producing fixed-length clips. Users can bind frequent actions to keys, address commands to the scene itself (flooding a room, rolling in a dust storm), start from a video instead of an image, and hand controls to agents that pursue objectives, with multiple people able to hold different roles in the same session. The launch matters because it extends the direction Runway set with GWM-1, the autoregressive General World Model built on Gen-4.5 that runs in real time and can be controlled through camera pose, robot commands, and audio , pushing from short controllable clips toward open-ended playable worlds. Runway was founded in 2018 by Cristóbal Valenzuela, Alejandro Matamala, and Anastasis Germanidis , ITP alumni from NYU , and GWM Worlds 2 is the clearest signal yet of the company's move beyond video generation into simulation as a general-purpose interface.