Apr 1, 2025

Firm says technology will produce consistent high-quality clips Cristobal Valenzuela, chief executive officer and co-founder of Runway AI. Photo: Kyle Grillot/Bloomberg Rachel Metz Today at 03:30 A new artificial intelligence model from Runway AI aims to let users create videos with consistent characters, objects and backgrounds, marking a possible leap ahead in the race to use computers to make films more quickly and inexpensively. The firm released Gen-4 to its paid users, with plans to add a function later in the week that is designed to make the software more adept at generating scenes that look consistent from one video to the next. Users will be able to generate clips that are five and 10 seconds long at 720p resolution, the company said. The New York-based start-up kicked off the frenzy around AI video generators in early 2023 with the release of a model that could produce slightly choppy-looking three-second clips based on written prompts such as "drone footage of a desert landscape". Two years later, the competition has increased significantly, with tech companies including OpenAI and Pika rolling out similar services. Now, Runway is vying to stay at the front of the pack. "Our goal right now is to make sure that the things that we're working on and building have the bar and quality that Hollywood and actual filmmakers and professionals require," said Cris Valenzuela, Runway co-founder and CEO. "It feels like a special moment because we've kind of crossed that threshold." Two short films made with the software that were shown to Bloomberg News offered a sense for what it can produce. In one clip, which was meant to resemble a fantastical claymation-like scene in the forest, a small, lonely campfire flame yearns for a connection and accidentally singes a skunk before finding a friend in a nearby rock. To make it, a member of Runway's team generated several hundred individual videos in a few hours, then edited them into a cohesive-looking clip. Sound was added separately. The entire process took a couple of days, Mr Valenzuela said. While AI videos have improved over the past two years, many still show all manner of inconsistencies, from phantom body parts to objects that defy physics and look noticeably different from one clip to the next. Mr Valenzuela said Runway's newest AI model is able to improve its outputs by retaining and sustaining a range of details, such as the location, characters, and the look and feel of the video. Mr Valenzuela declined to explain the technical details in depth, but said Runway had to rethink what its video model was aiming to do. Runway's software has been used for a range of projects, including generating some scenes for Amazon’s House of David, creating visuals for a Madonna concert tour, and making an ad for Puma. Runway also inked a deal with Lionsgate to train an AI model on the studio's content that can be used in its film projects. Mr Valenzuela said the company trained its model more on the jargon of film, in an effort to make the process of writing prompts more natural for the filmmakers using it. "The first goal was to render video, and I think that's kind of the state of video AI these days," Mr Valenzuela said. "The second stage is to actually make real stories – something people will watch because they're good and entertaining and you connect with."