The great royal portraits were never simply records of a face. They were arguments in oil paint.

A crown placed on a table suggested authority without requiring the sitter to wear it. A curtain opened onto an imaginary landscape. A diamond rivière caught light that may never have existed in the room. The monarch looked taller, calmer and more inevitable than any living person could manage for long.

Now those portraits are being asked to do something their painters never anticipated. With the help of artificial intelligence, a static figure can turn towards the viewer, change clothes, enter another room or appear to speak. A princess known only from one formal image can be rendered at breakfast, at a ball or walking through the palace garden. A vanished court can be rebuilt on a laptop.

The results range from enchanting to ridiculous. They also raise a surprisingly old question: when we look at a royal image, how much truth do we expect to see?

Royal portraiture was already a kind of image generation

Long before photography, few people would ever see their sovereign in person. The royal portrait did the travelling instead. Copies entered embassies, public buildings and the homes of loyal courtiers. Engravings carried the approved likeness further still.

These images were carefully constructed. Clothing announced rank. Jewels connected the sitter to a dynasty. Columns, thrones and rich textiles made power look permanent. Even a hand resting casually on a table could be part of the performance.

The tradition has not disappeared. The Coronation State Portraits of King Charles III and Queen Camilla place recognisable human sitters inside a visual language of robes, crowns and historic rooms. They record two individuals, but also present the institution those individuals represent.

In that sense, the court painter and the modern image model share a task. Both assemble a believable scene from learned conventions. Both decide what to emphasise and what to soften. The difference is that a painter makes those decisions consciously, while an AI system works through patterns in the images on which it was trained.

That difference matters more than it first appears.

The temptation to meet the person behind the pose

Royal watchers are used to reading details. The jewels can identify the occasion. A sash reveals an order. The cut of a sleeve can help date an otherwise mysterious photograph. Sometimes the smallest object provides the clue that connects one generation to another.

AI offers a seductive next step. What if the stiff formal portrait could relax, or a black-and-white court photograph could appear in colour?

An AI character generator can turn a description into a new fictional figure, allowing users to specify appearance, clothing and mood rather than simply selecting an existing image. Applied thoughtfully, the same principle can help visualise a court dress described in an inventory, test how a lost tiara might have sat on the head, or create a clearly labelled interpretation of someone for whom no informal likeness survives.

The key word is interpretation. A generated image may feel immediate, but it has not uncovered a hidden photograph. It is a new picture built from probabilities, references and somebody’s instructions.

Historical illustration has always involved informed imagination. The viewer simply needs to know where the evidence ends and invention begins.

The jewels are where AI gets tested

Faces are difficult, but royal jewels may be harder.

A tiara is not a generic crown. Its proportions matter, detachable elements may appear in one decade and vanish in the next, and many pieces were altered or reset. A model trained on loosely captioned images can easily combine two different objects into one convincing fantasy.

Consider the celebrated 1926 portrait of Queen Victoria Eugenie by Philip de László. The image is not memorable merely because a beautiful woman wears diamonds. The Fleur-de-Lys Tiara and the long diamond rivières have identities, histories and family connections. Replace them with a plausible-looking sparkle and much of the portrait’s meaning disappears.

AI reconstructions therefore require the same discipline as traditional research. A creator should begin with dated photographs, collection records and reliable descriptions. If part of a tiara is unknown, it is better to acknowledge the gap than allow software to invent a definitive answer.

The machine can produce an image in seconds. Knowing whether the image is right can take days.

Colour can clarify — and quietly mislead

Colourisation is one of the most popular ways to update historic photographs. When it works, the effect is immediate. A remote figure suddenly appears closer. Uniforms become legible, flowers separate from the background and a room begins to feel inhabited rather than archived.

Yet colour can also create false confidence. The dark fabric in a monochrome photograph might have been navy, green, burgundy or black. A gemstone that appears pale may be a diamond reflecting light or an aquamarine photographed under difficult conditions. Even when the colour of a dress is recorded in writing, the exact shade may remain uncertain.

AI tends to dislike uncertainty. Asked for an answer, it supplies one. The resulting blue sash or ruby necklace looks so natural that viewers may stop asking whether it is supported by evidence.

A responsible reconstruction can solve this with a caption. “Possible colour interpretation” is less dramatic than “Queen X in full colour”, but it tells the truth. Museums already distinguish between an original object, a restoration and a modern replica. Digital royal history needs the same vocabulary.

A moving face changes the emotional bargain

Animating a portrait is more complicated than adding colour. A blink seems harmless. A smile can feel touching. Once the figure begins speaking, however, the viewer is no longer merely studying an image. The dead appear to perform.

This can be powerful in education. A clearly marked reconstruction may encourage someone to learn about a historical figure previously known only as a name beneath a painting. It can also help younger audiences notice details of dress and setting.

But realism carries responsibility. A fabricated speech should not be presented as an authentic recording. A cloned voice should not turn a disputed opinion into a confident declaration. The more convincing the performance, the more clearly its invented elements must be identified.

Royal history is particularly vulnerable because it is already surrounded by myth. Animation adds another layer of interpretation — one that can spread widely after its original caption has been removed.

What happens when the reconstruction is wrong?

The internet rarely rewards caution. A beautiful image travels faster than its footnote, and a fictional portrait may soon appear on another account labelled as a newly discovered photograph. From there, it can enter family trees, fan pages and even poorly researched publications.

Royal imagery already suffers from mistaken identities. Cousins resembled one another, titles changed, photographs were copied without captions and formal dress concealed as much as it revealed. Generated material adds images that look archival without ever having passed through an archive.

There are practical ways to reduce confusion. Creators can mark the image, retain the source list and include “AI reconstruction” in the caption rather than burying it in a hashtag. Websites can keep original photographs beside the generated version so readers can compare them.

The goal is not to make experimentation joyless. It is to ensure that today’s creative exercise does not become tomorrow’s false evidence.

The royal double has political consequences

Historical figures cannot object to a digital reconstruction, but living royals can. A convincing image of a monarch at an invented event, wearing an unauthorised uniform or appearing to endorse a cause can move quickly through social media. During a constitutional or diplomatic crisis, the damage may occur before a palace communication team has time to respond.

Newsrooms, archives and fan communities all participate in the information chain. Viewers also need the habit of checking provenance: Who published the image? Is there another photograph from the occasion? Does the royal household or a reputable agency carry it?

Royal watchers are unusually well prepared for this. They already compare jewels, dates, guests and locations. The same attention used to distinguish two similar necklaces can expose an image that belongs to no real event.

The portrait will always be a performance

It is tempting to divide royal imagery into two neat categories: the truthful portrait and the artificial reconstruction. History is less tidy.

Painters flattered. Photographers retouched. Courts selected which images could circulate, while newspapers cropped them to tell different stories. Every official portrait reflects choices about costume, setting, posture and symbolism. The image has always been both document and performance.

AI does not create that tension. It makes the tension harder to ignore.

Used carelessly, generative tools will fill royal history with polished inventions and familiar faces wearing the wrong jewels. Used with curiosity and proper labels, they can help us test ideas, reconstruct lost appearances and notice details in original works that had faded into the background.

The most interesting result may not be the moment an old portrait appears to speak. It may be the moment the moving image sends us back to the still one — to check the tiara, read the date and look again at what the artist was trying to say.