Runway releases GWM Worlds 2 world model
Runway's GWM Worlds 2 produces 720p video with synchronized audio in real time and takes text commands as it runs, moving the company's world-model work toward games and simulation.
Runway has introduced GWM Worlds 2, the follow-up to the GWM Worlds environment-simulation model it released in December. The new version is built on the company's foundational audio-video generation model. It generates 720p video at 24 fps and audio at 48 kHz, and it keeps running from each new input, so sessions have no fixed length.
Users control the model through a format Runway calls WorldPrompt, which splits a world into two parts. The persistent part is a "genesis prompt" describing the scene, its subjects and the rules that govern them, from gravity to character abilities and camera perspective, plus a first frame for visual grounding. The second part is a timestamped stream of free-form text actions aimed at subjects or at the scene, along with per-frame camera input. Runway's examples include a sword slash, a spoken reply and a dust storm rolling in. Actions can overlap, and recurring ones can be bound to keys.
Runway says it first fine-tuned its bidirectional base model on WorldPrompt, then post-trained it into a real-time autoregressive diffusion model. Each step conditions on the global context, the current frame's inputs and recent frames held in a sliding KV-cache window. The video and audio decoders are causal. The model handles first- and third-person navigation, controls the camera and subject independently, and generates character dialogue with matching lip movement.
Runway names interactive entertainment, virtual characters, robotics and embodied-agent simulation as target uses. It notes that real-time control is the hardest mode, because typing and even VLMs are too slow to react within tens of milliseconds. Runway published no benchmarks and did not give availability or pricing.