From Video Generation to Simulation: Machine Intelligence for the Physical World
Runway's Anastasis Germanidis traces generative video from Gen-1 through Gen-4.5 and a real-time model, framing the progression as a move toward world simulation.
Event Runway AI Summit 2026
Speaker Anastasis Germanidis (Runway co-founder and co-CEO)
Released 2026-08-12 (YouTube)
Runtime 20 min 15 sec
Session Summary
Runway co-founder and co-CEO Anastasis Germanidis explains how media technology is moving from recording reality, through generating it, toward world models that simulate reality. This is an approximately 20-minute talk held at Runway AI Summit 2026.
The talk looks back at the evolution of Runway’s video models from Gen-1 through Gen-4.5. It then introduces the real-time version of Gen-4.5, developed with NVIDIA, and demonstrates interactive generation that returns the first frame within 100 milliseconds.
In user research conducted by Runway, fewer than 10% of participants could consistently distinguish generated videos from real videos. Germanidis connects this not merely to a video-quality metric, but to a progression in which models learn the rules of movement and space and develop into simulation tools.
The previous 24-minute listing differed from the actual video length of 20 minutes 15 seconds and has been corrected. The speaker’s name is shown in the session information rather than in the cast field.