Runway CEO: Gen 4 video model solves consistency, enabling solo filmmakers to produce short films in days

Summary

Runway's Gen 4 video model, launched the day of this conversation, targets the single biggest obstacle in AI video: consistency. CEO Cristóbal Valenzuela describes it as the ability to build coherent worlds across shots — the prerequisite for actual storytelling rather than isolated clips.

The clearest proof point is a short film called The Lonely Little Flame, produced entirely with Gen 4 by a single person. Valenzuela says the first rough cut took two to three days, with a few more days to finalize. Production value of that kind, he argues, would have cost hundreds of thousands of dollars through traditional means.

How Gen 4 works in practice

The model lets filmmakers combine reference images, color grades, and locations without relying on text prompts. Valenzuela is direct that prompting is overrated as an interface for video — language can't capture composition, mood, or style with enough precision. Gen 4 also supports motion-tracked performance capture, where a creator's own movement drives the generated output, effectively making the creator their own storyboard.

A second short, The Herd, demonstrates another capability: starting from a single image and repositioning the camera across a scene. Valenzuela describes the model behaving like a cinematographer, surfacing angles and ideas a director couldn't afford to explore on a physical set.

The Hollywood cost problem

Hollywood budgets have become self-defeating. When a production costs hundreds of millions of dollars, studios default to franchises and superheroes — the bets they know will return. That economics has crowded out independent stories. Valenzuela argues Gen 4 inverts that logic: low enough cost that one person can make something with real production value, which removes the studio as a necessary gatekeeper.

On whether studios are responding fast enough, he notes Disney CEO Bob Iger called AI the most important technology for his company just days before this conversation. Valenzuela says many studio partners are aligned, though some are further along than others.

The quality benchmark is shifting

Valenzuela wants to retire photorealism as the eval for AI video. The Will Smith eating spaghetti clip was the old benchmark; Gen 4 is past it. The internal goal is to be judged by story quality and audience engagement — awards, viewership — not by whether the image looks real. The uncanny valley that took VFX twenty years to cross, he says, AI video is crossing in two or three.