
Capacity Expo
How We Brought Napoleon back to Life
The creators of the new exhibition in Slavkov u Brna came up with an idea that sounded simple: What if visitors could talk to Napoleon? We’re not talking about a pre-recorded video or a paid actor. What if they could actually talk to an avatar that thinks, speaks, and possesses the same knowledge, desires, and motivations as the emperor himself? We accepted this challenge and embarked on one of the most technically demanding projects we’ve ever undertaken.
Project Vision
Have a real audience with the emperor. An experience where you feel as though you’re standing face to face with one of the greatest (not in terms of height) Frenchmen, and he’s talking to you. He has his own personality and accent, and he treats you like a subject, not a museum visitor. And of course—he answers your questions.
Challenge
Personality: Create a personality for Napoleon based on knowledge drawn from scholarly books, letters, and museum collections. At the same time, it should be prepared to handle provocations and inappropriate questions.
Accent: Give him a slight French accent in every language—one that sounds natural and human, yet can navigate all the pitfalls of artificial intelligence.
Real-Time Movement: Our Napoleon moves in real time. This isn’t a pre-recorded video or a complicated animation that will bore you to death because it’s fifteen seconds behind.

We didn’t approach the project as a technological implementation, but as a directorial task. We devised the concept of “An Audience with the Emperor.” Visitors do not arrive as tourists with a guide, but as citizens who are to cherish this honor. And Napoleon treats them accordingly—as an emperor treats his subjects—with authority, distinctive humor, temperament, and pride.
Napoleon’s persona was developed in close collaboration with historians and the authors of the libretto and exhibition concept—his manner of expression, rhetoric, emotional patterns, and relationships with people and events. To this end, we compiled our own database of sources and historical documents. Thanks to this, Napoleon is able to answer specific, unexpected, and even intimate questions about his life.
What the visitor perceives as a fluid conversation is, however, a series of five interconnected AI models running in real time.
1. The visitor's voice is converted to text
2. Napoleon's response is generated
3. A voice with an accent is generated
4. Emotion is added
5. The speaking face is animated


Services that generate synthetic voices are designed to produce perfect results, with the voiced characters speaking clearly and neutrally. But what if the replicated voice is supposed to have a clearly recognizable accent? After all, Napoleon without a French accent isn’t Napoleon. We had to work meticulously to ensure he had a slight French accent in all languages.
That’s why we manually adjusted the voice models until we found the perfect voice. Napoleon now has a French accent in Czech, English, and German. At the same time, the voice isn’t flat—it has its own complexity. When he reminisces about love, he sounds melancholic; when he talks about the Russian campaign, he sounds angry. Just as a real person would.
Historical depictions of Napoleon exist almost exclusively in profile, so the front view had to be calculated and reconstructed. All of this was done to ensure that it wasn’t just a static portrait.
The result is a lifelike image in which the entire body moves—not just the mouth. We created loops of natural movements, such as hand gestures and head movements. These run as an organic layer beneath the precisely synchronized facial animation. The result doesn’t look like a photograph brought to life, but like a real person.
To ensure the best possible experience for visitors, we focused on how they would interact with Napoleon. We knew they would try to throw him off balance, outwit him, or corner him. They would come up with inappropriate questions, provocations, or simply talk nonsense. During the preparation phase, we optimized the experience with the understanding that people might try to trick the model in some way.
And the result? Napoleon doesn’t come across as confused or robotic. He brushes off inappropriate questions with imperial grace. He dismisses uninteresting questions, stating that no one will waste his precious time. He ends the conversation with such charm that the visitor leaves with a smile, even if they were turned away.
The Napoleon AI project for the new exhibition in Slavkov wasn’t just an attraction for visitors to the castle. Thanks to targeted media outreach—which we also took under our wing—it became a topic that resonated across national, technology, specialized, and tourism media. The exhibition was covered by Seznam Zprávy, Novinky.cz, CzechCrunch, MF Dnes, Metro, Deník N, TV Nova, Česká televize, Český rozhlas, Archiweb, Earch.cz and many others.
A particular highlight was the report on TV Nova’s main news program and the in-depth article in Seznam Zprávy, which significantly expanded the topic’s reach beyond the usual cultural and regional context. Overall, according to media monitoring, the communication efforts in the first month after the opening generated over 65 media appearances with a potential reach of over 105 GRP, or approximately 10 million consumers.

Result
Napoleon works. He speaks, gestures, reminisces, and even gets angry. In short, he chats with anyone who steps in front of him. The entire system operates in real time, without relying on external paid services and without any noticeable delay.
It’s an installation that proves that the best technology is the kind that doesn’t even seem like technology. Like, for example, chatting with the French emperor.
