Now loading...
New York, December 11, 2025 – AI firm Runway unveiled GWM-1 on Thursday, describing it as its most advanced general world model to date, designed to replicate and interact with real-world scenarios in real time.
The model family, built upon Runway’s Gen-4.5 foundation, operates as an autoregressive system that produces video sequences frame by frame. It enables interactive control through inputs such as camera movements, robotic instructions and spoken audio, allowing users to guide simulations dynamically.
GWM-1 includes three specialized versions: GWM Worlds, which creates navigable virtual spaces; GWM Avatars, focused on lifelike digital characters for dialogue; and GWM Robotics, aimed at aiding mechanical tasks. Currently, these function as distinct post-trained systems, but Runway plans to integrate them into a unified framework covering diverse actions and settings.
Runway first outlined its general world models initiative two years ago, positioning these AI tools as mechanisms that construct internal maps of surroundings to forecast outcomes. The company argues that such models represent a pivotal advancement in artificial intelligence, surpassing limitations of text-based systems for tackling complex issues in areas like automation, medicine and research.
By facilitating rapid simulations, GWM-1 allows virtual trial-and-error learning akin to human processes, but at accelerated speeds without physical risks. Key capabilities include conditioning on actions like visual shifts, occurrences, mechanical positions and voice inputs, alongside adaptability for custom applications and support for videos lasting up to two minutes.
In robotics, GWM Robotics acts as a digital emulator to produce training data, bypassing hardware constraints and enabling safe policy assessments. It forecasts video progressions based on machine directives and handles hypothetical scenarios. A new Python software development kit provides access to the model’s application programming interface, supporting multi-angle views and extended timelines for integration with contemporary robotic frameworks.
GWM Worlds offers instantaneous rendering of expansive, coherent realms from initial scenes, preserving spatial logic during exploration and incorporating user-defined physical rules. This supports applications in video games, learning programs, agent preparation and virtual reality setups.
For GWM Avatars, the system drives character animations via sound cues, capturing authentic facial cues, gaze tracking, mouth synchronization and body language in prolonged interactions. It suits uses like virtual instruction, client assistance and entertainment, with rollout planned soon via Runway’s online platform and developer interface.
Alongside GWM-1, Runway enhanced its Gen-4.5 toolset with built-in sound creation and modification features, permitting fresh video clips with accompanying audio or alterations to existing ones. A multi-shot editing function also lets changes in starting frames carry over consistently across full productions, regardless of length.
These developments underscore Runway’s push toward versatile AI simulations that bridge digital and physical domains. For more on the announcement, visit Runway’s research page. Details on Gen-4.5 are available at its dedicated section.
