Runway has released Gen-4, its latest AI model for video generation. The company says the system addresses one of the biggest challenges in AI video generation: maintaining consistent characters and objects throughout scenes.
According to Runway, Gen-4 creates more dynamic videos with realistic movements while keeping motifs, objects and styles consistent across frames. "Gen-4 sets a new standard for video generation and is a marked improvement over Gen-3 Alpha," Runway states.
New approach to character consistency
The key technical advance is Gen-4's ability to maintain consistent characters using just one reference image across different lighting, locations and treatments. It combines visual references with text instructions to preserve styles throughout videos.
According to Runway, Gen-4 also represents "a significant milestone in the ability of visual generative models to simulate real world physics." The system allows users to place any object or subject in different locations while maintaining consistency.
Runway created several demo films to showcase these capabilities. "New York is a Zoo" demonstrates character consistency by placing the same animals into different New York scenes, combining animal reference images with city photographs.
"The Herd" shows a chase scene through a nighttime cow field, with Gen-4 maintaining consistent character appearances using only a few reference images.
"The Retrieval," showing explorers searching for a mysterious flower, was created with Gen-4 in under a week.
The Gen-4 image-to-video model is available to paying subscribers and Enterprise customers. Reference functions will be added in a future update. Additional examples are available on RunwayML's website.
RunwayML has established itself as a major player in AI video technology, with a valuation of around $4 billion, reported annualized revenue of $80 million in December 2024, and projected annualized revenue of $300 million in 2025. It's the strongest Western competitor to OpenAI's Sora.
