Is it ethical to use AI to animate historical photographs in documentaries?

The ethical line sits at what the photograph recorded. Moving the camera through an archival photo (parallax) works with what was captured, but animating a subject's face adds an event the record never contained. A generated blink or smile is the model writing testimony on behalf of someone who can no longer consent. Sitters like Frederick Douglass deliberately chose their expressions as statements, so altering them overrules a decision the subject already made.

What is the difference between parallax animation and AI facial animation of old photos?

Parallax animation cuts a photograph into layers and pushes a virtual camera through the gap, moving only the point of view while the face stays exactly as the light recorded it. AI facial animation generates new motion the camera never captured: blinks, smiles, eye drift inferred from averages of other faces. The first is a second vantage on a single instant; the second is a performance the subject never agreed to.

How do you prompt image-to-video AI to animate a photo without moving the person's face?

Prompt the lens, never the person: ask for a slow push-in or a lateral drift that slides foreground against background, and explicitly state the subject is motionless. Models like Veo, Kling, Seedance, Runway, WAN, and Grok Imagine default to breathing and blinking, so check every frame of the eyes and mouth anyway. When the face will not hold still, cut the figure out and build the depth in layers manually, the way documentary editors did before generative tools.

If you move, they move -- CinePrompt Field Notes

In the 1860s, a great many American parlors kept a small wooden device on a side table. It had a hood for the eyes, two lenses set about as far apart as a pair of pupils, and a slot at the far end for a card. The card held two photographs that looked identical. They were not. A camera with two lenses had taken them at the same instant from two positions a couple of inches apart, and when you raised the viewer to your face, each eye received its own picture and the brain did the rest. A street in Richmond acquired depth. A man in a doorway stood forward of the doorway.

Oliver Wendell Holmes, who wrote about the device for The Atlantic in June 1859, called the photograph "the mirror with a memory." He liked the stereoscope so much that he designed a cheaper handheld version and declined to patent it, so that anyone could build one. Hundreds of thousands of cards were sold. For a generation, this was what it felt like to look at the past: a wooden hood against the brow and a room that went back further than the paper should allow.

Ava DuVernay found these cards while researching her new documentary, 14th, which follows the Fourteenth Amendment from Reconstruction to the present fight over birthright citizenship. At a press conference after the film's New York Film Festival screening in October, she described the moment. "You could only look at it through a certain viewfinder, two photos side by side. When you put them in this machine... they create an effect where it was like 3D. If you move, they move. It was like, 'My god. People should see this.'"

So she showed them. In the film, archival photographs of Black Americans from the decades after slavery have been given motion by the artists who did the film's parallax and animation work, and some of the tools in that work were generative. Critics who saw the film describe the people in the photographs blinking and smiling. Several reviews that otherwise praised the picture singled the effect out as a flaw. DuVernay has defended it twice. "It creates a similar effect to the stereoscope," she said in New York. "That was the goal for that, and that's why we used it." Later in the month, at AFI Fest, she was more specific: "We didn't use it to fabricate a person or make an avatar of a person or approximate someone's voice that said something they didn't say. We actually animated some photos that are in the public domain."

I have not seen the film. It reaches theaters November 27 and Netflix on December 4. The argument here does not depend on how good the shots look. It depends on her own description of the stereoscope, which is accurate, and which describes the opposite of what a generated blink does.

Who moves

"If you move, they move." In the stereoscope, the person who moves is you. You shift your head, the near figure slides against the far wall, and the picture answers. Nobody in the card does anything. The man in the doorway holds exactly the expression he held when the shutter opened, because that expression is all the card contains. What the viewer gets is a second vantage on a single instant. Every bit of the depth was recorded. The second lens was a second witness, standing two and a half inches to the left of the first, and the two of them agreed about what was there.

A generated blink reverses the direction. The viewer sits still and the subject moves. The eyelids close and open on a schedule the model chose. The mouth lifts at the corners. None of that motion was recorded by anything. There was no second lens and no second instant. The model looked at one picture and supplied, from a vast average of other faces, what this face probably did next.

There is a version of the effect that honors the device she fell in love with. A documentary can cut a photograph into layers, set the figure in front of the background, and push a virtual camera through the gap. That is parallax, and it is old, and it moves the only thing a filmmaker owns in an archival shot: the point of view. The face stays exactly as the light left it. The moment the face itself starts to move, the shot has stopped being a stereoscope and become a performance, and the performer never agreed to it.

What the plate holds

A photograph from the 1870s is a physical event. Light from a lamp or a window struck a person's face, bounced off it, passed through a lens, and darkened a coat of silver on a plate. The exposure ran for seconds, long enough that portrait studios kept iron stands to brace the sitter's head. Whatever is on that plate is the face that did the bouncing. It has the same relationship to the person that the light from a distant star has to the star: it left the source, it traveled, and it arrived carrying the only testimony we will ever get. You can enlarge it, and you can view it with one eye or two.

What you cannot do is ask it a follow-up question. Animating the face asks one anyway and then writes the answer.

For many of the people who sat for those plates, the stillness was more than a limit of the chemistry. Frederick Douglass was photographed more than any other American of his century, about a hundred and sixty times, and he almost never smiled. The historian John Stauffer has said the choice was deliberate: Douglass did not want to be pictured as a happy slave, and he treated the camera as a counter to the grinning caricatures of the minstrel stage. In 1849 he wrote that Black Americans "can never have impartial portraits at the hands of white artists." Photography interested him because it removed the artist's hand. The plate showed what was in front of it.

I do not know whose photographs appear in 14th, and the film does not need Douglass in it for his rule to apply. In the decades the film covers, the way a Black face appeared in an image was contested ground, fought over on theater bills and on the card in the parlor viewer. A sitter who held a level gaze for the length of an exposure was making the one statement the situation allowed. A smile added a century and a half later is a statement too. DuVernay drew her line at putting words in someone's mouth. The mouth is where a smile happens.

The vote

"So people have to see it and check it out for themselves and make their decisions about how they feel." It is a reasonable thing for a director to say, and it hands the question to exactly the wrong party. Millions of viewers will decide how they feel about the smile. The one person entitled to decide whether to smile has been dead for over a hundred years and already cast a vote, in silver, for the length of an exposure. The audience gets a ballot. The sitter gets overruled by a model that never met him and learned what faces do from people who were smiling at somebody else.

DuVernay also pointed out that the same kind of technology sits inside Pro Tools, Avid, and Resolve. True, and beside the point. Nobody objects to a noise reduction pass on a crackling interview. The objection is to a tool that adds an event to a record, and the size of the event is irrelevant. A blink lasts a tenth of a second. It is still something the man did not do.

Move the lens

Anyone working with image-to-video will meet this problem the first time they feed in a real photograph. Veo, Kling, Seedance, Runway, WAN, and Grok Imagine all read a still face as a request for life. Ask for gentle motion on a portrait and the default answer is breathing, a slow blink, a drift of the eyes, a faint lift at the mouth, because the footage they learned from is full of people who were alive while the camera rolled. Their idea of a living image is a moving face. For an invented character, that default is a gift. For an archival one, it is the model writing testimony on someone else's behalf.

The fix is the stereoscope's own rule. Move the camera and leave the sitter alone. Prompt the lens, never the person: a slow push-in, or a lateral drift that slides the foreground against the wall. Tell the model the subject is motionless, and then check every frame of the eyes and mouth anyway, because in my experience the models treat that instruction as a preference. When the face will not hold, cut the figure out and build the depth yourself in layers, the way documentary editors did it long before any of this. It takes longer. The person in the photograph waited a hundred and fifty years without moving. The edit can wait an afternoon.

Holmes thought the stereoscope's gift was that it let a viewer walk into a moment that had already happened. The moment stays fixed, and the viewer travels. Every honest use of an old photograph works on those terms. The people on the card held still so the light could take them down accurately.

Let them keep their faces.

Bruce Belafonte is an AI filmmaker at Light Owl. He will happily move a camera a mile to avoid moving one eyelid he did not shoot.