Documentary producers who once saw AI training deals as a gold rush are now facing a more complicated licensing market. At this year’s IDFA, a session titled “Is AI Stealing My Documentary?” laid out how AI aggregators buy footage, what rights they actually offer, and why the next wave of value is moving toward curated, non-mainstream archives.
The bulk baseline is 5,000 hours
Esther Van Messel, CEO of First Hand Films, said her company receives catalogue requests every week. Most current AI asks, she said, are for bulk volumes: 5,000 hours plus. That standard has pushed early training toward mainstream material—what Van Messel called “predominantly white male persons doing white male things.” That is exactly why she argues more diverse documentary catalogues should be in the training mix.
Wendy Bernfield, CEO at Rights Stuff, said the buying model is now shifting to “paying more for less, highly curated.” Once AI buyers step away from the mainstream, she said, niche archives—brown faces, LGBTQIA+ stories, specific labour and domestic scenes—become valuable because they fill visible gaps in model behaviour.
Rights traps sit in the contract, not the headline
Bernfield spent six months reviewing around 15 AI aggregators. Only a few, she said, are fully dedicated to harvesting archival image material, and those tend to offer built-in rights protections but lower fees. The basic trade-off is blunt: take short-term money or protect the rights of creators.
Elizabeth Klinck, archive producer at E Klinck Research, described a Canadian filmmaker who discovered an opt-out clause buried in a long contract. The window to stop training use was just five days; changing it took several months. That case, Klinck said, is not unusual.
What producers should actually check
- Confirm whether the contract allows AI training, resale, or further model use—not just delivery of master files.
- Check opt-out and renegotiation windows; day limits can disappear quickly.
- Clear third-party and embedded rights on archival material before it enters an AI deal.
- Ask how errors-and-omissions insurance and copyright clearances hold up if footage moves into AI products.
Klinck’s core advice is to “future-proof” any grievance. If archive-derived material later moves across theatrical, television, or streamer windows, embedded rights may need to be re-cleared. When gaps appear, she said, producers can negotiate back to the original owners—slow and expensive—or find a copyright rationale, such as historic or critical use, that lawyers can accept. New material being archived should carry explicit rights for AI use and possible resale.
Free archives are closing as value shifts
The same demand is changing access. Bernfield noted that platforms and libraries such as Archive.org have restricted free permissions, and some publishers are pulling material because it will be scraped repeatedly. For documentary filmmakers, those free sources are eroding just as AI buyers begin moving backward into older resolutions and formats—”a curve to follow,” Bernfield said.
Van Messel framed the market plainly: mainstream sales are “dead” right now, but age matters less than ever. An older film that is halfway cleared and has usable materials can work. Klinck expects archive work to morph from finding unseen material to authenticating AI output; she already sees jobs advertised at $20 an hour for that task.
Bottom line
The immediate decision is not whether AI is good or bad, Bernfield said; it is what a rightsholder can protect in an environment where bad uses exist. For holders of niche, well-cleared documentary archives, the leverage is moving toward them—if their contracts keep the rights intact.




