Artificial intelligence start-ups are willing to pay substantial sums for film and television IP, but what are the factors library owners should consider before taking up a potential ‘once-in-a-generation’ opportunity? The subject will be among the hot topics at Content London next week.

John Wick maker Lionsgate has expeirenced AI controversy firsthand

When John Wick, The Hunger Games and Mad Men maker Lionsgate unveiled a “first-of-its-kind partnership” with New York-headquartered artificial intelligence (AI) start-up Runway, neither was likely expecting the backlash that followed. But given we’re in an age of social media, surely some was anticipated considering the sensitivity of the topic?

“I wonder how the directors and actors of their films feel about having their work fed into the AI to make a proprietary model?” Reid Southern, a concept artist on the original Hunger Games movie, posted on X. “This is the first step in trying to replace artists and filmmakers.”

“Our words, performances and direction are merely to feed the machine until we’re no longer needed,” wrote actor, director and writer Alexander Chard.

The timing of the announcement didn’t help, coming as it did just shy of the anniversary of the end of the US writers’ strike, which, together with the actors’ action, brought Hollywood to a halt for six months last year. At the heart of both the Writers Guild of America (WGA) and Screen Actors Guild (SAG-AFTRA) disputes were concerns over how studios (represented by the Alliance of Motion Picture & Television Producers) might use AI to cut costs, freeze artists out of the creative process or squeeze further profits from work already done without offering additional compensation.

Michael Burns

And lo, less than 12 months later, here was Lionsgate heralding an alliance with a company that, together with fellow purveyors Stability AI and Midjourney, remains subject to a class action lawsuit from a group of artists alleging copyright infringement.

“Letting the fox into the hen house,” was how one Tinseltown agent, who spoke to C21 on condition of anonymity, described the deal. “OK, great, you’re going to use AI to make production better. But as an agent, my first question is how is that helping my client? Are my clients going to get paid more money? Are you going to pass those costs on to us? None of the creatives I talk to are sure how it works yet,  and most of the talent community are afraid of AI replacing them.”

Neither Lionsgate nor Runway would comment for this article beyond the joint statement issued at the time in which Lionsgate vice chair Michael Burns described the tech firm as “a visionary, best-in-class partner who will help us utilise AI to develop cutting-edge, capital-efficient content creation opportunities.”

“Several of our filmmakers are already excited about its potential applications to their pre-production and post-production process. We view AI as a great tool for augmenting, enhancing and supplementing our current operations,” added Burns.

“As we continue our work with Lionsgate and other studios, we’re also considering ways to license or otherwise offer these models as templates that can be made available to individual creators interested in building and training their own proprietary models,” said Runway.

There are a couple of crucial points to note here: Runway’s focus is on building ‘proprietary models,’ meaning the firm can’t, in principle, use Lionsgate’s IP or that of any other partner for its own purposes. It cannot produce generative AI (GAI) content of its own – based on, for example, the John Wick franchise – but internally, Lionsgate could do so in areas such as storyboarding or the design and creation of special effects. Critically, the terms of the SAG-AFTRA agreement reached last November would preclude it from replicating star Keanu Reeves, or indeed any other actor members.

“We are aware of the Lionsgate/Runway AI arrangement. We will ensure that the use of GAI in that arrangement, as it relates to the work of our covered performers, complies with SAG-AFTRA terms and existing law,” Duncan Crabtree-Ireland, SAG-AFTRA’s national executive director and chief negotiator, told C21.

Terminator was a cautionary tale about the dangers AI presents to humanity

Also, Runway says it is working with other studios, and in the fast-developing field of AI it is by no means alone. Meta, Google and Microsoft – backer of ChatGPT and Sora creator OpenAI – are all said to be in talks with Hollywood, eager to license vast treasure troves of film and TV to help train their respective AI systems. And, reputedly, willing to pay large amounts for the privilege.

Steve Hulford is co-founder and CEO of Underknown, a Toronto-based digital media company that owns and operates popular YouTube channels including What If, Aperture, How to Survive and Animalogic, the latter a nature and wildlife outlet acquired this year from Blue Ant Media.

The serial entrepreneur recently spied a new business opportunity after online tech journal 404 Media reported possessing a leaked document detailing how Runway had trained its text-to-video tools on a string of YouTube channels run by the likes of Disney, Netflix and National Geographic, with Underknown’s own Animalogic also on the list of sources.

“Runway scraped thousands of YouTube channels; that’s where its learning came from. How do we know this? 404 Media wrote an article about this not even a month or two ago,” says Hulford. “YouTube has since launched a feature where we can opt out of being scraped, which Underknown has chosen to do.”

But what this revelation prompted him to explore was not the possibility of a lawsuit, rather the licensing opportunities that exist among AI protagonists prepared to play by the rules.

In the publishing sector, OpenAI has already struck licensing deals with the likes of Associated Press, Axel Springer, the Financial Times and News Corp – the latter worth US$250m over the next five years, according to The Wall Street Journal – while a pact between Google and Reddit hands the latter US$60m a year, says Reuters.

Sora has set the industry on edge

Sora

These arrangements relate to text, so OpenAI uses them to train ChatGPT, and likewise Google with Gemini, but it is the former’s text-to-video AI tool Sora, unveiled to shock and awe in February, that stands to have the greatest impact on the film and TV industries. But arguably, it first needs the latter’s assets to achieve its full potential.

Bloomberg reported earlier this year that Meta, Google, Microsoft and the like are willing to pay “tens of millions of dollars” to Hollywood studios for the right to ingest their content.

Lionsgate has a catalogue of more than 20,000 film and TV titles – small in comparison to majors Warner Bros Discovery (200,000-plus hours), Paramount (140,000-plus), NBCUniversal (110,000-plus) and Disney (100,000-plus), none of which were willing to comment for this story. By comparison, in Europe, Banijay claims a portfolio of some 190,000 hours, whereas ITV Studios cites 90,000 and BBC Studios and Fremantle both boast upwards of 40,000. These companies, too, either declined to give interviews or did not respond to requests for them.

AI is a beast that requires sizeable data sets to feed it, but Underknown’s Hulford is trying to act as an intermediary, teaming up with smaller library owners to aggregate programme packages of sufficient scale to be of interest, not only to the big-name AI players but a host of other start-ups too.

“It’s a once-in-a-generation opportunity where you’ve got these AI models that are hungry for content,” he says. “We’re working with a partner in the US to take advantage of this, to round up hundreds of thousands of hours of content, and it could be substantial sums of money for television production companies.”

Steve Hulford

How much exactly? “Suffice to say, someone with 3,000 hours could make between £300,000 [US$400,000] and £600,000 in 2025. It’s a new opportunity so the range is highly variable. The needs of the AI engines are different as well. We have around 40-plus targets for this content,” says Hulford

He has now set up a dedicated AI licensing business within Underknown called Krill.

“The AI models are very, very hungry and the most voracious animal to ever have swum the earth is the blue whale. And what does the blue whale eat? It eats krill, so these AI models are the blue whale in this analogy and video is the krill,” he explains, adding that he is looking for rights holders with over 400 hours of content.

Underknown is by no means the only operative emerging in this space. Austin, Texas-based Troveo, which is led by creator economy veteran Marty Pesis, allows content owners to upload their content to its platform and do licensing deals with AI companies looking to train their video models. According to the firm, rights holders have the ability to license their content an unlimited number of times, depending on how many distinct platforms wish to use the content for training.

“AI companies have been seen as adversaries to the creative industry, but we believe there’s a better path forward. We’re enabling creators to work side-by-side with AI companies in a way that ensures they are fairly compensated for their contributions,” says Pesis, who recently hired former TikTok and CBC executive Gave Lindo as Troveo chief content officer.

Meanwhile, Los Angeles-headquartered Calliope Networks was founded last year by NBCUniversal, Paramount and 20th Century Fox exec Dave Davis and data scientist Jim Golden, after similarly seeing an opportunity in the fact that many emerging AI text-to-video tools are feeding off online copyrighted content, with rights holders seeing no reward.

“In many cases, they are scraping content from the internet and they’ve taken a position that they have the legal right to do that, which we would dispute,” says Davis. “We’re tracking more than 80 companies that are in the AI video world. Some of those are really unwilling to engage with us but most of the largest are interested in the conversation.”

Underknown bought Animalogic from Blue Ant Media

Due to non-disclosure agreements, Davis is unable to say which Calliope is in talks with, but he is frank about the scale of the financial reward available to those film and TV rights holders willing and able to license their catalogues accordingly.

“Our list price for HD content is US$6.25 per minute,” he says. “When you have a chaotic market, you end up with the risk of having no market at all because expectations are so disparate from the buyer and the seller, so we view one of our roles as really making a market and establishing a set of principles and prices that is going to be acceptable to both sides.

By this metric, the library of a distributor like Banijay, which claims a portfolio of some 190,000 hours, would be worth over US$70m, if it were all in HD (higher premiums are available for 4K). But the exec says things aren’t quite as black and white. “If you were licensing 190,000 hours, there would probably be some volume discounts associated but there is real value to that catalogue,” he adds.

Elsewhere, Mariano Rosón, CEO of US Hispanic production company Nitro Group, is in the market for libraries of TV shows, movies and podcasts that can then be licensed to AI companies trying to develop things like virtual psychologists or virtual language teachers.

“The tech companies foresee that not far in the future there will be legislation making it very simple for a company to claim somebody is using their rights and get a subpoena to search all their servers,” says Rosón. “A little further ahead, that will be automatic. In the same way we can find if somebody is using our content online, this will happen with servers and everything will be discovered. So acquiring content under legal terms is the most sensible way to do it.”

Caitlin McGivern

Six months after Google announced its US$1.65bn acquisition of YouTube in October 2006, Viacom sued the online video-sharing start-up for US$1bn, alleging massive copyright infringement. In response, Google introduced YouTube Content ID to automatically detect infringements in user-uploaded videos. Viacom’s lawsuit rumbled on for seven years, during which Content ID evolved and media companies became more comfortable with the platform.

“It’s worth thinking about how YouTube, Napster and other disruptors worked. They knew they were breaching copyright and people didn’t quite know how to deal with it. Is it worth me suing someone for a tiny little clip that just appeared on their website?” reflects one prominent UK TV exec who preferred not to be named.

“There is an old adage that copyright is technology’s child and whenever there’s a technological leap, there’s always a lag whilst governments and laws catch up and then we find a new normal. The AI companies are not going to stop. They are disruptors backed by the disruptors from 10, 20 years ago. We need to find our way through it, but obviously, the one thing they really need to truly fine-tune their models is our content, so we hold the power, to a degree.”

The exec is referring to the likes of Stability AI, the UK-headquartered start-up behind GAI tool Stable Diffusion that appeared on the brink of bankruptcy earlier this year but was rescued in June by investors including former Google CEO Eric Schmidt and notorious Napster founder Sean Parker. The latter, now executive chairman, even welcomed James Cameron to the company’s board in September, despite the Titanic and Avatar director having once warned his 1984 hit Terminator was a cautionary tale about the dangers AI presents to humanity.

“The YouTube/Content ID analogy is a good one,” says Gary Woolf, executive VP of strategic development at All3Media International, the distribution arm of Redbird IMI-owned UK production group All3Media, which holds a catalogue of over 30,000 hours.

“We can certainly see that there are monetisation opportunities in this space. We’re at an early stage in a handful of conversations and what’s really important to us, in terms of representing our producers and rights holders, is that the content going into these training models is ethically sourced – essentially, licensed. As to what the terms are, we’re just in the middle of unpicking that.”

“These studios and IP owners have so much content they own the rights to, it does make sense to try to encourage people to dip into that rather than into unlicensed content,” says Caitlin McGivern, a senior associate at UK law firm Harbottle & Lewis.

For McGivern, the Lionsgate/Runway deal was a “savvy decision.” She says: “From a legal and practical perspective, it’s a question of Lionsgate being the first to put their hand up and say, ‘Look, we have to get out ahead of this.’”

She acknowledges it is a sensitive topic, especially among creatives concerned for their jobs. But she sees the deal as an attempt to give those with whom the studio works the AI tools they can use to advance their craft using legally cleared source material rather than undermine them, and “reduce or remove the risk of infringement claims resulting from it.”

Guillaume Pommier

Many questions remain, of course, not least the extent to which, despite WGA and SAG-AFTRA ‘guardrails’ and emerging legislation in the US, Europe and elsewhere, some countries and companies may adopt a more relaxed attitude to IP. A number of Sora rivals have sprung up in China, notably ByteDance’s Jimeng AI, Shengshu AI’s Vidu and Kuaishou Technology’s Kling AI.

“If we were to be approached by someone who’s willing to license our content, we would be super careful about where the company is based and what the interests of its shareholders are,” says Guillaume Pommier, co-head of distribution at Federation Studios, producer of biographical drama series Bardot.

Not only that, but the exec is also wary about the longevity of AI start-ups, given the space is so nascent and hyped.

“None of these AI companies are profitable right now, so it’s kind of a bubble,” he says. “OpenAI is worth billions (US$157bn to be precise after closing a new US$6.6bn round of funding), but how long will they last and what happens if you give them exclusivity but they close?”

Pommier also has environmental concerns, given the incredible energy demands of AI data centres. Microsoft, for example, recently struck a deal to help restart the Three Mile Island nuclear plant – scene of the US’s worst nuclear accident in 1979 – for such purposes.

The immediate issue for a lot of TV companies in 2024, however, is that their production income has tanked.

“Since the peak TV era, it’s been very hard for everyone to finance shows,” Pommier continues. “But in the meantime, it’s the era of the library. Your library can help you finance future development, so we need to be super cautious about  how we exploit that, take care of it and get the best out of the IP for the next 30 years.”

Hear more from Steve Hulford, Dave Davis, Gary Woolf and Guillaume Pommier in the C21Podcast tomorrow and don’t miss their panel discussion together with Harbottle & Lewis partner Rupam Davé at Content London on Thursday 5 December at 4pm in Hall 1 of Kings Place.