AI-Animated Famous Portraits: Marie Curie, Van Gogh, Napoleon

The Musée d'Orsay holds a self-portrait by Van Gogh. Oil on canvas, 1889. A fixed gaze, tense features, closed mouth.

A few months ago, a team of artists animated that portrait. Van Gogh turns his head slightly. His eyes move. The tension in his features relaxes for a fraction of a second.

Ten seconds of video. One hundred and thirty-seven years disappear.

These portraits that nothing had ever made move

The idea of animating famous historical faces is not new. It has just become accessible.

The first viral wave dates to 2021, with MyHeritage's Deep Nostalgia. Millions of people uploaded ancestor photos. Many also tried it with historical figures: Lincoln, Napoleon, pharaohs. The results were decent but mechanical. Every face made the same head movement, the same eye blink. An animated mask, not a living face.

What changed is the underlying technology. Today's video diffusion models don't overlay a pre-recorded movement onto an image. They analyze bone structure, eye angle, lighting, posture, and generate movement that is coherent with that specific face. Each animation is unique. Van Gogh does not move like Lincoln. Marie Curie does not move like your great-grandmother.

That is why current results produce a different effect than those of 2021. Less the wonder at a gadget, more something harder to name: recognition.

Marie Curie, Einstein, Napoleon: History's portraits in motion

Since 2022, examples have multiplied. Brazilian artist Hidreley Diao created photorealistic portraits of historical figures from paintings: Julius Caesar, Alexander the Great, Cleopatra, George Washington. Museum teams have integrated animated portraits into temporary exhibitions to capture younger audiences.

Marie Curie is one of the most frequently animated figures, probably because several photographs of her exist with decent resolution for the era. Her 1920 portrait produces a particularly sharp result: she turns her head slightly, the eyes move, the expression stays the same — concentrated, almost impatient.

Albert Einstein, Charles Darwin, Van Gogh, Abraham Lincoln: their animated portraits have been circulating for several years. The effect is consistent. People watch the video twice. Then three times. Then they look for other portraits to animate.

This is not nostalgia. It is something more precise: the discovery that a face can cross time without losing what makes it recognizable.

How AI animates a static portrait

The principle is simple to grasp, hard to execute.

Models are trained on millions of videos of human faces. They learn how eyes move relative to eyebrows, how the neck turns with the shoulders, how facial muscles deform during a smile or a blink. Not hand-coded rules: patterns learned through massive observation.

When the model receives a static photo, it uses these patterns to predict how that particular face would move over time. The bone structure under the cheekbones, the angle of the eye sockets, the tension of the jaw: all of this informs the generated movement.

First-generation models applied fixed movement sequences. Seedance 1.5 Pro, the model powering Incarn, generates unique movement for each image. That is not a detail: it is the difference between an animated mask and a living face.

Our article on AI photo animation technology goes deeper on the technical approaches if you want to explore further.

What this is, and what it is not

A clarification before continuing.

Animating Marie Curie's portrait does not produce a recording of how she actually moved. The AI generates plausible movement, coherent with the visible facial structure in the photo. It is an interpolation, not a historical reconstruction. A machine-assisted interpretation, in the same way a biography is not a live broadcast.

Cultural institutions working with this technology make this clear. The Rijksmuseum has integrated animated portraits into some exhibitions, always accompanied by a note explaining the process. This honest framing is part of the respect owed to the people depicted.

Some find the effect unsettling. The uncanny valley — that discomfort in front of a representation that is almost human but not quite. It usually fades by the second or third animation, as the eye adjusts. That is a normal response to something new. It is not a reason not to try.

For broader ethical questions — rights, consent, appropriate use of faces — our article on AI photo animation ethics covers the nuances in detail.

Your ancestor deserves the same treatment

This is where everything becomes personal.

Famous portraits have something most family photos do not: high resolution, professional framing, careful restoration. But they do not have what you have: the knowledge of the name, the life, the history of the person depicted.

A user wrote to us after animating a photo of his mother, who died in 1987. He watched her move for the first time in thirty-eight years. He told us he had cried, and that his grandchildren had finally "seen" their great-grandmother breathe.

Napoleon, Marie Curie, Van Gogh: their portraits are in museums. Your great-grandmother is in a cardboard box in your attic. She deserves just as much to live.

More than 12,000 photos have been animated on Incarn since launch. One at a time, at $1.99 per photo. The first one is free.

For damaged, scratched or faded old portraits, our guide on restoring old photos with AI explains how to improve quality before animating.

Frequently asked questions

Can AI animate any historical portrait?

Animation works well when the face is visible and relatively sharp. Hyperrealistic paintings, good-quality daguerreotypes, early 20th-century photos with decent lighting all produce good results. Frescoes, medallion profiles, and very blurry portraits are more difficult. Practical rule: if you can clearly recognize the face in the original, the AI can work with it.

Do museums actually use this technology?

Yes. Several European and American institutions have experimented with animated portraits in temporary exhibitions or digital installations. Use remains primarily educational, always accompanied by a note about the AI origin of the movement.

Can a low-quality photo be animated?

Prior restoration significantly improves results. Tools like Remini, Adobe Photoshop (Enhance), or Topaz Photo AI reduce noise and improve sharpness before uploading. In practice: a photo scanned at 300 DPI minimum, without major damage on the face, produces the best results.

Why do some people find it "unsettling"?

The uncanny valley. The discomfort some people feel in front of a representation that is almost human but not quite. It usually fades after a few viewings, as the eye adjusts. It is also related to seeing a familiar — or well-known — face in a state never observed before. A normal response. Not a reason not to try.