Runway, one of several AI startups developing video-generating tech, today announced an API to allow devs and organizations to build the company’s generative AI models into third-party platforms, apps and services.
Currently in limited access (there’s a waitlist), the Runway API only offers a single model to choose from — Gen-3 Alpha Turbo, a faster but less capable version of Runway’s flagship, Gen-3 Alpha — and two plans, Build (which is aimed at individuals and teams) and Enterprise. Base pricing is one cent per credit (1 second of video costs 5 credits), and Runway says that “trusted strategic partners” including marketing group Omnicom are already using the API.
The Runway API also comes with unusual disclosure requirements. Any interfaces using the API must “prominently display” a “Powered by Runway” banner linking to Runway’s website, the company writes in a blog post, to “[help] users understand the technology behind [applications] while adhering to our usage terms.”
Runway, which is backed by investors including Salesforce, Google, and Nvidia and was last valued at $1.5 billion, faces stiff competition in the video generation space, including from OpenAI, Google and Adobe. OpenAI is expected to release its video generation model, Sora, in some form this fall, while startups like Luma Labs continue to refine their technologies.
With the preliminary launch of the Runway API, Runway becomes one of the first AI vendors to offer a video generation model through an API. But while the API might help the company along its path to profitability (or at least recouping the high costs of training and running models), it won’t resolve the lingering legal questions around those models and generative AI tech more broadly.
Runway’s video-generating models, like all video-generating models, were trained on a vast number of examples of videos to “learn” the patterns in these videos to generate new footage. Where did the training data come from? Runway refuses to say, like many vendors these days — partly out of fear of sacrificing competitive advantage.
But training details are also a potential source of IP-related lawsuits if Runway trained on copyrighted data without permission. There’s evidence that it did, in fact — a report from 404 Media in July exposed a Runway spreadsheet of training data that included links to YouTube channels belonging to Netflix, Disney, Rockstar Games, and creators like Linus Tech Tips and MKBHD.
It’s unclear whether Runway ultimately ended up sourcing any of the videos in the spreadsheet to train its video models. In an interview with TechCrunch in June, Runway co-founder Anastasis Germanidis would only say that the company uses “curated, internal datasets” for model training. But if it did, it wouldn’t be the only AI vendor playing fast and loose with copyright laws.
Earlier this year, OpenAI CTO Mira Murati didn’t outright deny that Sora was trained on YouTube content. And Nvidia reportedly used YouTube videos to train an internal video-generating model called Cosmos.
Many generative AI vendors believe that the doctrine known as fair use provides legal cover — and they’re asserting this in court and in public statements. Others are less inclined to take chances, and/or they view a more “ethical” approach to model training as a selling point for their services. To develop its video-generating Firefly models, Adobe is said to be offering artists payments in exchange for clips, for example.
However the lawsuits pertaining to the legality of training on copyright content shake out, one thing’s becoming clear: Generative AI video tools threaten to upend the film and TV industry as we know it. A 2024 study commissioned by the Animation Guild, a union representing Hollywood animators and cartoonists, found that 75% of film production companies that have adopted AI have reduced, consolidated or eliminated jobs after incorporating the tech. The study also estimates that by 2026, more than 100,000 of U.S. entertainment jobs will be disrupted by generative AI.
