How Runway's CEO thinks about AI video, creativity, and the next media format

Executive overview

AI video tools have crossed an inflection point — models can now modify existing footage in ways that were impossible a year ago. Runway's CEO Cristóbal Valenzuela argues this isn't just a productivity tool for Hollywood; it's the beginning of an entirely new media format that sits beyond both film and games.

The opportunity is to treat AI as a new medium rather than a substitute for existing craft. Studios, architects, and marketers are already using Runway in pre- and post-production. The company resets its roadmap weekly and declined a Meta acquisition to stay independent.

AI video is not a shortcut — it's a new medium that rewards those who learn to think with it, not just prompt it.

What makes Runway's approach distinct

• Built on nearly eight years of focused development before competitors entered the market
• Gen 4 model enables video modification using existing footage — not just generation from scratch
• Target users are professionals who want creative control, not just fast output
• Single underlying model can serve architects, game studios, filmmakers, and marketers without vertical-specific products

How Hollywood actually uses AI video

• Pre-production: storyboarding, script development, character exploration, location visualisation
• Post-production: applying visual effects, changing lighting conditions, correcting scene continuity
• Classic example: turning a day scene into night avoids reshooting entirely
• Studios treat it as a cost and time reducer, not a film generator
• AI in film is no longer taboo — exposure has replaced initial resistance

The new medium argument

• Every major technology breakthrough has produced a corresponding breakthrough in art
• Painters misjudged photography; today's critics repeat the same error with AI video
• Linear media (film) and nonlinear media (games) are both top-down designed worlds
• Real-time AI generation produces worlds with no pre-scripted rules — a genuinely different category
• Valenzuela: "We don't have the language to describe that thing yet"

On operating in a fast-moving environment

• Product roadmap resets weekly; quarterly planning is too slow to survive
• Research and product must be integrated — pure research misses customer feedback, pure product gets leapfrogged
• Runway uses open-ended research boundaries rather than fixed goals, letting teams discover unexpected value
• Speed of model improvement, compounding performance gains, and near-frictionless distribution create zero-to-billion timelines never seen before

On responsibility and deepfakes

• Valenzuela frames it as collective social responsibility, not solely a company problem
• Analogy: cars required seatbelts and licences — society builds norms around powerful technology
• Clear disclosure context (watching a film vs. a Zoom call) is how the distinction will be maintained
• Exposure to the technology is what shifts public understanding and builds appropriate norms

On the Meta acquisition approach and staying independent

• Meta reportedly approached Runway before its $14B Scale AI deal
• Valenzuela declined: too invested in the vision, wants to own the outcome
• Believes Runway has a rare ability to extrapolate where model performance is heading
• Views independence as necessary to pursue the long-term new-media format thesis

Expansion beyond video

• AI video models generalise to robotics and autonomous vehicle training data
• Generating synthetic training data (e.g. 10,000 lighting variations) replaces expensive real-world data collection
• "Everything that will be moving pixels on a screen will probably be generated with AI very soon"