Runway releases Gen-4

Runway released Gen-4, an update to its video-generation model aimed at a limitation that had confined AI video to short, disconnected clips: consistency. Given a single reference image, Gen-4 could generate a character, object or location that stayed visually coherent — the same face, the same outfit, the same room — across separate generations shot from different angles, in different lighting, or in different settings, rather than each new clip effectively reinventing its subject.

Runway described this as combining reference images with text instructions to produce new images and video that preserved consistent styles and subjects across multiple shots, and said the model represented a step forward in simulating real-world physics — motion, weight, and interaction between objects — compared with its predecessor. Multi-angle coverage of the same scene, a basic requirement of conventional filmmaking that earlier generative video tools struggled to deliver, became achievable without re-describing the subject from scratch each time.

The release mattered less for a single benchmark than for narrowing the practical gap between AI video tools and the production workflows of film and advertising, where continuity across shots is a baseline requirement rather than a bonus feature. Runway, competing against OpenAI’s Sora, Google’s Veo and a crowded field of Chinese video generators, positioned Gen-4 as evidence that consistency — not just visual fidelity in isolated clips — was becoming the frontier that mattered for commercial adoption of generative video.