Runway drops Gen-4.5 image-to-video: only 10% can spot the fakes

Beyond the naked eye

Runway just rolled out its new image-to-video capabilities for Gen-4.5, pushing AI-generated realism to an unnerving new level.

The 4.5 generation was already sitting comfortably at the top of the Artificial Analysis text-to-video leaderboard. But the real headline was an internal blind test Runway shared alongside the release: only 10% of participants could tell the difference between real footage and videos generated by the model.

Runway itself called this milestone a societal "tipping point." Low-grade AI clips are already duping viewers across social feeds, but models like Gen-4.5 push visual fidelity to a point where artifacts are practically undetectable.

Why it matters

We think this release signals the definitive end of "seeing is believing."

For video creators, animators, and filmmakers, Gen-4.5 is a serious creative leap forward. Turning a still image into a convincing, cinematic shot is now accessible with minimal friction, radically cutting pre-production and VFX costs.

The darker side, of course, is verification. Even as the tech industry pushes for provenance standards and watermarks, Runway admits there is still no silver bullet for a world where people can no longer trust their eyes. When 90% of viewers fail a basic sniff test, the defensive playbook looks dangerously behind the curve.