Almost everyone who has watched a lot of documentaries in recent years has experienced the moment when something in a frame doesn’t sit quite right. The light is too uniform. The movement of a crowd loops almost imperceptibly. A half-second is too long for a face to maintain its expression. You shake it off, blink, and continue to observe. However, the doubt has already surfaced and is persistent. It turns out that the quiet uneasiness is not paranoia. It is the perfectly reasonable reaction of a viewer gradually realizing that the agreement it previously had with documentary filmmakers has been subtly altered.
The basic, unstated premise of documentaries has always been that what you see is real. There was a camera. It was pointed at a real object. The underlying photographic record acted as an anchor, but that promise was never absolute because editing, framing, and narration all distort reality. That anchor has been raised by generative AI. And the film industry is still standing on the deck, watching it sink, at least the portion of it that makes nonfiction work.
“People will stop believing anything. In both cases, our regime of truth is completely shaken.” — Susana de Sousa Dias, IDFA 2025
The discussions backstage and in panel rooms at last year’s International Documentary Film Festival Amsterdam, the biggest event in the world devoted to nonfiction film, kept returning to the same fear. The festival’s Guest of Honor, Portuguese filmmaker Susana de Sousa Dias, stated bluntly that there is more danger than just audiences believing fake archival footage. It’s the opposite risk that could be more serious.
People will cease to believe anything. She pointed out that both results undermine the shared regime of truth on which documentaries have always relied. It’s the kind of observation that sticks with you. Suddenly, a century of visual evidence—wars, natural disasters, political upheavals, and human faces caught in moments of pure, unguarded emotion—is obscured.
David France, an Oscar-nominated filmmaker, saw this coming before most. While filming Welcome to Chechnya in 2019, he encountered a truly unfeasible circumstance: LGBTQ individuals who were escaping persecution in Chechnya, the subjects of his documentary, were being pursued. They put their lives in danger by revealing their faces. His team digitally superimposed the faces of volunteer queer activists on those of the documentary’s subjects using machine learning, which they weren’t yet referring to as artificial intelligence. It was successful.
It was so outstanding that it was recognized with a Technical Academy Award. However, France was accused of creating deepfakes right away. “Deepfake is the crime, AI is the tool” was his insightful and memorable response. Even though the general public has mostly stopped making this distinction, it’s still very important.
One of the things that makes navigating the current moment so challenging is the blurring of that distinction. Was it a manipulation of reality or a compassionate restoration of it when France later employed AI to revitalize and restore the voice of Leonard Peltier, an activist who had been imprisoned for fifty years and whose recordings had been obtained under difficult circumstances? There isn’t a simple solution. Transparency, not deceit, was the goal. However, an audience that is unable to see past the edit cannot perceive the intention.
Instead of avoiding the ambiguity, British director Marc Isaacs chose to embrace it. His most recent film, Synthetic Sincerity, which debuted at IDFA, examines what happens when you work with AI to develop a character that exists in between generated and real. Isaacs collaborated with Romanian actress Ilinca Manolache to create real emotional range, which he believed AI couldn’t provide on its own. He experimented with Synthesia, a platform for synthetic media, but discovered that its AI capabilities were limited. dead.
He centered the project on a human performer while simulating AI-generated aesthetics with filters and techniques. He didn’t want to “spoil” the main question of the movie, which is why he refused to identify what was and wasn’t synthetic. That is an artistic decision that can be justified. The question of whether it’s a responsible documentary choice is a whole other story.
Perhaps the most difficult of all is the archival problem. Grainy, low-definition, black-and-white footage from the past has always had an air of incompleteness that suggested authenticity. These antiquated artifacts served as a sort of evidence in and of themselves. The visual grammar of authenticity can now be faked thanks to AI that can produce extremely lifelike historical imagery. It is possible to create “archival” footage in the aesthetic register that conveys the idea that “this is real.” This was brought to IDFA’s attention by Sousa Dias, and the implications are extremely serious. The kind of historical memory that documentaries have spent a century creating might become truly unverifiable.
Rob Horning made a crucial observation in his article for Aperture magazine: the term “real photo” has lost its meaning. When we say it, we mean that we think someone with a point of view stood somewhere and recorded something, not that an image has some verifiable ontological status. Because a generative image appears incorrect, it does not violate our sense of reality. It appears correct most of the time. It violates it because it deceives us into giving something that lacks a human gaze, consciousness, or witness. That is the more profound loss. The implied presence behind the image is just as important as the image itself.
Where this goes for the documentary genre overall is still unknown. As alternative venues for nonfiction storytelling, some filmmakers are turning to virtual reality and interactive online spaces, where audiences have developed different types of trust and the environment’s constructed nature is already evident. One intriguing signal is the Netflix documentary The Remarkable Life of Ibelin, which takes the virtual world of an online game seriously as a place where actual human life was truly lived. Perhaps a documentary is necessary if reality is partially shifting into virtual environments. It’s an odd turn of events, but it makes sense.
What is certain is that documentary filmmaking is about to enter a phase in which every decision—whether to employ AI restoration, create illustrative images, or recreate the aesthetics of a bygone era—will need to be revealed, discussed, and defended in front of an increasingly skeptical and right-to-be audience. Technology isn’t really the key to the genre’s survival. It concerns whether filmmakers can restore the confidence that artificial imagery has started to erode. It turns out that the camera was more than just a recording tool. It served as a witness. Additionally, witnesses are capable of lying, as anyone who has witnessed a trial is aware. Now, the question is whether the documentary will be able to remind viewers that even the best of them refuse to.
