Runway's GWM-1 and Gen-4.5 Enable Live-Steerable AI Video, Targeting Sub-100ms First Frame on Nvidia Vera Rubin

Runway is developing real-time AI video generation that allows users to steer video output live, frame by frame, moving beyond the current prompt-and-wait workflow. This advancement, built on Runway's General World Model (GWM-1) and Gen-4.5 video model, was first previewed with Runway Characters in March and aims to deliver the first frame in under 100 milliseconds on Nvidia's Vera Rubin platform.

Shifting from Prompt-and-Wait to Live Steering

The core innovation in Runway's development is the transition from a static, batch-processed video generation workflow to an interactive, live-steerable system. Instead of submitting a prompt and waiting for a complete video, users would be able to stream video content as it generates, providing real-time input to guide the output. This method allows for dynamic adjustments and creative control during the generation process, moving closer to an experience akin to live editing or interactive simulation.

This capability is built upon Runway's foundational models: GWM-1 and Gen-4.5. GWM-1, introduced in December 2025, is designed to accept various control inputs, including camera movements, robotic commands, or audio cues, enabling a broader range of interactive applications. The integration of these models is crucial for achieving the responsiveness required for live steering.

Technical Foundations and Performance Targets

Runway's real-time video generation leverages its General World Model (GWM-1) and the Gen-4.5 video model. The company has also addressed visual inconsistencies by training its models on their own flawed outputs, a technique designed to enable self-correction and improve output quality over time. This iterative self-improvement mechanism is vital for maintaining visual coherence in a live generation environment.

A research preview, demonstrated at GTC on Nvidia's Vera Rubin platform, showcased the ambition to deliver the first frame of generated video in under 100 milliseconds. This performance target is critical for achieving a truly real-time and interactive user experience, minimizing latency between user input and visual feedback. The shift to real-time generation also reallocates the computational burden from the training phase to the inference phase, which could potentially lead to reduced costs per output.

Implications for Interactive Applications

The development of live-steerable AI video generation holds significant implications for various interactive use cases. Industries such as gaming, education, and robotics stand to benefit from the ability to generate and manipulate video content in real time. For instance, in gaming, this could enable dynamic, procedurally generated environments that respond instantly to player actions. In robotics, it could facilitate the evaluation of robotaxis or other autonomous systems through real-time visual simulations.

While Runway first previewed this real-time approach with Runway Characters in March, the company has not yet announced a specific availability date for the broader real-time video generation capabilities. This ongoing research and development aim to transform how users interact with AI video generation tools, making the creative process more fluid and immediate.

Conclusion

Runway is advancing real-time AI video generation through its GWM-1 and Gen-4.5 models, moving towards a system where users can live-steer video output. This initiative, demonstrated with performance targets like sub-100ms first frame delivery on Nvidia's Vera Rubin platform, aims to unlock new possibilities for interactive applications in fields ranging from gaming to robotics. While the technology is still under development, it represents a notable step towards more dynamic and responsive AI-powered creative tools.

Sources

• Runway started by helping filmmakers — now it wants to...
• Runway releases its first world model, adds native audio to latest video model | TechCrunch
• Runway Research | Introducing General World Models
• Runway rolls out Gen 4.5 AI video model that beats Google, OpenAI
• Runway wants to turn AI video generation into a live stream you control in real time