How Multimodal generative AI Reads the DNA of 500 Years of Art

Have you ever wondered how the chaotic, messy evolution of human creativity might share the same underlying elegance as a biological genome? Nature organises life through DNA, a highly structured code hidden beneath a wild variety of living things. Human culture seems much more random. Paintings from the Renaissance look completely different from modern art. Yet, scientists are now finding a hidden code within our art history.

Multimodal generative AI meets the canvas

To find this hidden structure, researchers turned to computers. They used Multimodal generative AI, specifically a programme called Stable Diffusion, to look at 500 years of Western paintings. Most people use these tools to make new images. Here, the team used the technology in reverse. They wanted to see how the computer organises human art in its digital brain. They asked the AI to extract two types of data from the paintings. First, they looked at formal details, like colours and brushstrokes. Second, they looked at contextual details, which means the actual subjects and themes of the artwork.

The evolution of our cultural genome

This brings us to a fascinating philosophical detour. Why would human culture evolve with a specific, measurable structure? In biology, a genome is organised to help an organism survive in a changing environment. Traits adapt to new threats and opportunities. Human art might do exactly the same thing. Art evolves to reflect the shifting pressures of society, politics, and technology. When the researchers measured the AI data, they found something surprising. The contextual details—what the paintings were actually about—matched up perfectly with known artistic periods and individual artists. The formal details, like colour, did not line up nearly as well. This suggests that the true genetic code of an artistic movement is not found in the paint itself. Instead, it lives in the ideas and subjects the artist chose to explore. The AI could actually track how artistic expression shifted alongside major historical events.

Painting the future of the past

The team did not stop at simply measuring the data. They ran an experiment. They took historical paintings and asked the AI to inject them with ideas from different time periods. The computer successfully created new images that perfectly matched the style of those target eras. This study measured how an AI categorises visual and thematic data. What it suggests is far bigger. It implies we can now map the hidden structure of cultural knowledge. By treating art history like a giant, evolving organism, we can use modern computing to read its genetic code.