Entertainment

What Happens When Documentary Makers Sell Their Footage to AI Aggregators?

What Happens When Documentary Makers Sell Their Footage to AI Aggregators?

Specialists at IDFA discussed the dos-and-dont's of licensing material to AI training machines as the practice becomes more common among rights holders

Back in 2024, leading documentary film festival IDFA opened with Piotr Winiewicz’s “About a Hero,” a documentary that set out to test the premise of whether or not an AI model could conceive a Werner Herzog film. Back then, the film received some backlash for its use of the technology, with docmakers— including Herzog —openly rebuffing the use of artificial intelligence in cinema.

Like the technology itself, much has changed in terms of discussions around AI in the two years since, with more and more industry experts admitting to not only using AI in their films but also willingly feeding the training systems. Recently, lauded filmmaker Ava DuVernay admitted to using AI in Netflix doc “14th” to animate public-domain archival photographs and bring historical periods to life. Related Stories 'Wolfenstein 3' in Development at Xbox's MachineGames Need the Day Off to Play 'GTA 6'? Xbox and DJ Khaled's 'Dr. Xbox' Have a PTO Excuse Note for You

This year, IDFA faced the conversation head on with an industry talk aptly titled “Is AI Stealing My Documentary?” Wendy Bernfield, CEO at Rights Stuff, bluntly told the audience how the reality is that, for more than three years, AI has been “scraping or stealing before we were even aware enough about it.”

“It’s almost like the early days of YouTube,” she noted. “AI is following that path. For the first few years, anything on the web at all—that means your trailers or ad hoc channels, fast channels—was simply taken; you could argue for training so that they can train the robots to handle the model practice.”

Bernfield noted how, less than two years ago, AI branched into AI aggregation. These are platforms that combine output from several AI models into a unified interface. “What I did at the beginning was spend six months interviewing all different aggregators and looking at what they put in their contracts,” she said. “It wasn’t about asking permission and paying money, but also what rights, if any, are built in to protect you if you decide to license your contract.”

Out of the around 15 aggregators currently available, only a few have a full-time dedication to harvesting archival image material. Those few have built-in rights for protection, but that also means their fees are lower than riskier models rights-wise. “You have to decide if you’re in it for the short-term money gain or if you’re also trying to protect your rights and those of your creators,” she said.

Archive producer at E Klinck Research, Elizabeth Klinck said she is approached “just about every week” by filmmakers who have had AI parties enquiring about how to acquire their material. Since the field is still quite new and there is still a lack of dedicated roles to clear the legalities of such processes, Klinck, a veteran archive producer, became many docmakers’ first point of contact.

The producer recalled a Canadian filmmaker who found out “an opt-out clause to not have her material used for training” in a lengthy contract. “There was a five-day limit on that opt-out clause. And so in the end it took several months to get that changed because she really didn’t want the material used for training.”

“I realized there were a lot of people who were in that same boat that maybe didn’t realize some of the contracts they were signing had implications for their own work,” she added.

The first of many precious pieces of advice Klinck offered during the talk was to “future-proof” any grievances. “If you’re using material that is derived from archive, it may not hold up in terms of your errors and omissions insurance. It may not hold up in terms of your copyright clearances.”

This is because conversations around copyright can shift depending on what’s on the table, from theatrical to television and streamers, and also third-party rights, given that archival material often comes with embedded rights that often needs to be re-cleared. “The real practical pathways when all those gaps are discovered are that you can negotiate and go back to the original owners,” she said. “That’s sometimes time-consuming and sometimes expensive. But you can also figure out if there’s a way you can find a copyright section to use that material. So if it’s historic or if it is a creative, critical approach, then sometimes you can get the lawyers to agree that it can be used.”

“Finally, just make sure any new material you’re bringing in as archived has very specific rights that make it allowable for you to use AI and then possibly sell your material,” she noted.

CEO of First Hand Films Esther Van Messel also gets requests for their catalogs every week. At the beginning, she said, she was “very impressed.” “I thought: I can make money for my filmmakers, and this is going to be a gold rush.”

Van Messel pointed out that most of AI requests at the moment are for bulk, which means 5,000 hours plus, “and that means [it has to be] mainstream. So everything that AIs are training on right now is predominantly white male persons doing white male things.” This is one of the key reasons why the producer is very open to selling her material to AI: “I think our films should train AI. I love watching these reels with the AI actors. They can’t load a dishwasher. Obviously this AI model has never been trained on a very non-male, non-wide task like loading a dishwasher. So I think they should train with our content especially, but of course, within the limits and boundaries.”

On the issue of volume, Bernfield said the AI-buying model is currently shifting to “paying more for less, highly curated.” “So this is where your stuff will become more interesting because now it’s like, ‘Hey, wait a minute. We don’t have any brown faces. We don’t have any LGBTQIA+.’ You know, when they go off the mainstream, suddenly the niche becomes important.”

Another shift is how, in the beginning, AI only cared about “1080P, HD.” “Archives would not be relevant for an AI at that time,” noted Bernfield. “But now that they’ve got their HD108OP type high def, then they’re moving backwards into the format. It’s a curve to follow.”

“I mean, the bad news is the market is dead,” said Van Messel. “Nobody’s buying anything anymore. But the good news is there’s no old films because nobody cares how old something is. You can work with archives; you can work with the older films. If they’re halfway cleared and have halfway decent materials, we can work with them.”

Conversely, AI’s interest in archival material is leading precious research platforms and libraries to shut down free access to their resources. Bernfield mentioned how U.S. platforms like Archive.org have restricted their free permissions, with many libraries doing the same. “Places like The Guardian are saying, ‘We’re not gonna put our stuff there because we know it’s gonna be scraped over and over and over again.’ And so those great places that everybody could go to as a documentary filmmaker and not have to pay to get excellent material are now being really eroded.”

“I think it’s a threat to archives for sure,” said Klinck. “I think the job is going to really morph, and I think a lot of material is not going to be authentic or won’t have integrity. I think that’s going to be a real issue. And I think the job that we have as people who find unseen material will disappear and will be turned into a police kind of authenticating. I already see jobs on LinkedIn. They want to hire someone like myself for $20 an hour to authenticate the new AI.”

When the conversation shifted into dealing with the issues of selling footage of subjects disclosing sensitive information or personal trauma and the massive consequences AI has to the environment, Bernfield said these are questions entering the territory of “is AI bad?”

“We all agree AI can be bad,” she said. “In the environment where they’re doing bad things, what can you do as a rights holder or a producer to minimize or protect [against] one part of the theft, even if you can’t change the entire world?”


Source: Entertainment

More in Entertainment →