On viewFilm & Screen

AI aggregators seek documentary footage as archive access narrows

IDFA producers face AI licensing offers, restrictive contracts and shrinking free access as documentary archives weigh bulk deals against creator rights and source integrity.

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

Programme

  1. IDFA opened its 2024 edition with Piotr Winiewicz’s About a Hero, which tested whether AI could conceive a Werner Herzog film.
  2. Rights Stuff’s Wendy Bernfield said only a few of roughly 15 AI aggregators include archival-rights protections.
  3. One training opt-out gave a Canadian filmmaker five days to act and took several months to change.
  4. Most AI requests seek bulk licenses of 5,000 hours or more, Esther Van Messel said.
  5. LinkedIn already lists AI-authentication work at $20 an hour, Elizabeth Klinck said.

Only a few of roughly 15 AI aggregators include archival-rights protections, Wendy Bernfield told IDFA.

At the panel “Is AI Stealing My Documentary?”, Bernfield, CEO of Rights Stuff, said AI had been “scraping or stealing before we were even aware enough about it” for more than three years. She compared that phase with YouTube, when material across the web was simply taken, often with model training offered as the justification.

IDFA opened its 2024 edition with Piotr Winiewicz’s About a Hero, a documentary that tested whether an AI model could conceive a Werner Herzog film.

The film drew backlash. Herzog and other documentary makers openly rebuffed AI in cinema. Two years later, industry experts were openly using AI and feeding training systems. Ava DuVernay said she had used AI in the Netflix documentary 14th to animate public-domain archival photographs and bring historical periods to life.

Elizabeth Klinck, archive producer at E Klinck Research, said filmmakers approach her about AI licensing almost every week. The field lacks dedicated legal-clearance roles, she said, so she has become many documentary makers’ first point of contact.

The buyers

Bernfield spent six months interviewing aggregators and reviewing their contracts for permission, payment and protections for rights holders. An aggregator, as she defined it, combines output from several AI models within one interface.

Only a few of the roughly 15 services currently available devote full-time effort to harvesting archival images, and their built-in protections make them safer than riskier models. Those services charge lower fees. Bernfield said producers must choose between short-term income and longer-term protection for themselves and their creators.

Contract traps

A Canadian filmmaker found an opt-out clause for training in a lengthy contract, Klinck said. It allowed five days to act. Changing the clause took several months because the filmmaker did not want her material used for training.

Klinck advised documentary makers to future-proof potential grievances. Material derived from archives may not hold up under errors-and-omissions insurance or copyright clearances, she said. Many filmmakers may not realize how contractual terms can affect later uses of their work.

Klinck said archival material often carries embedded rights that require clearance again. Obligations can change between theatrical, television and streaming releases, as well as deals involving third parties.

Her practical options were to negotiate with original owners, which can be time-consuming and expensive, or identify another copyright route for using the material. She said lawyers may agree when a historic, creative or critical approach applies. New archive material needs specific rights covering AI use and any later sale.

A selective market

Esther Van Messel, CEO of First Hand Films, also receives catalog requests every week. At first, she said, she saw a way to earn money for filmmakers and expected a gold rush. Most requests now seek bulk licenses of 5,000 hours or more, and such purchases require mainstream material with wide recognizability in the training set.

Van Messel said the resulting data disproportionately represents white men performing what she described as white male activities. She remains open to licensed sales. She cited AI actors’ inability to load a dishwasher to argue that training data omits many everyday tasks. Her position has limits: filmmakers’ content should train AI only within agreed boundaries.

Bernfield said the buying model is now “paying more for less, highly curated.” As buyers move beyond the mainstream, they will seek brown faces and LGBTQIA+ subjects, she said, making niche work more valuable. The technical appetite has also reversed. Early systems wanted 1080P HD and ignored archives. Newer systems are moving back toward archival formats, Bernfield said.

Van Messel offered a harsher assessment of the market. She said buyers were no longer purchasing anything and the market was dead. She also said age now carries less penalty, opening possibilities for older films that retain usable quality and partial rights clearance.

The cost of scarcity

Bernfield said AI demand has already narrowed free access. U.S. platform Archive.org has restricted free permissions, and many libraries are following suit. She said The Guardian has declined to place material on such services because it expects repeated scraping. Free collections that once gave documentary makers affordable access are eroding.

Klinck called AI a direct threat to archives. She expects the curator’s role to change as generated material enters collections, with some material lacking authenticity or integrity. She said jobs once centered on finding unseen material will shift toward authenticating new AI output. She cited LinkedIn listings seeking such expertise at $20 an hour.

The panel also addressed footage involving disclosed trauma or sensitive personal information, along with AI’s environmental costs. Bernfield said participants agreed that AI can be bad, but rights holders can still minimize one part of theft in the transactions they control.

What does this change for documentary producers?

Producers should read training clauses before signing, map every embedded right and specify whether a buyer may train on or resell the material. Safer aggregators charge less; bulk offers may bring more money but carry weaker protections.

The next licensing cycle will reward documented provenance and curated collections. Unclear contracts and reduced free access will leave producers carrying more risk unless they resolve rights before footage leaves the archive.

Original: indiewire.com

documentary-film, artificial-intelligence, archival-footage, copyright-licensing, ai-training-data

Also on view

« Previous articleNext article »