Runway used NVIDIA’s GTC conference this week to demonstrate something that shifts the ground beneath AI video production: real-time video generation. A research preview model running on NVIDIA’s new Vera Rubin hardware produced HD video with time-to-first-frame under 100 milliseconds. That’s not a faster version of the current workflow. It’s a different workflow entirely.
To put that in context: current AI video generation is a batch process. You write a prompt, submit it, wait, and get a clip back. It’s closer to sending a brief to a post house than it is to directing. What Runway showed at GTC is generation that responds as you go – frame by frame, in real time, reacting to inputs as they happen. The company’s CTO, Anastasis Germanidis, presented the underlying architecture, GWM-1 (General World Model), which generates video autoregressively rather than in batches, maintaining physics, geometry, and spatial consistency as a user navigates or controls the scene.
The practical implications land in three places. First, previsualisation. If a director or creative lead can see AI-generated footage responding in real time to their inputs – camera moves, lighting changes, scene adjustments – the gap between concept and execution collapses. You’re not waiting for renders. You’re directing. Second, interactive content. GWM-1’s Worlds variant creates explorable environments; its Avatars variant produces conversational characters with real-time lip sync and expression. These aren’t demos looking for a use case. Advertising, training, and immersive formats are already there waiting. Third, cost. Real-time generation on purpose-built hardware means fewer cycles, less compute time, and ultimately cheaper production per minute of usable output.
Runway’s Gen-4.5, currently the top-rated video generation model on the Video Arena leaderboard, was ported to Vera Rubin in a single day, which tells you something about NVIDIA’s backward compatibility and something about how quickly this infrastructure can be adopted by studios already in the Runway ecosystem.
This is still a research preview. It isn’t shipping to subscribers tomorrow. But the direction is unmistakable: AI video generation is moving from “submit and wait” to “direct in real time.” That changes the creative process, the production timeline, and eventually the economics of everything built on top of it.
