The Architecture of Sfumato

Kelsey Song, Sichang Xu, Xinlu Sun

A{AI} - Architecture and Artificial Intelligence


Matias del Campo + Sandra Manninger


We start with a simple question: Mona Lisa is causing congestion in Louvre due to her popularity. Thousands of people come to visit Louvre just to take a picture of this famous painting. This not only causes congestion in the exhibition space of Mona Lisa, but also makes Louvre become more and more crowded. We decided to take Mona Lisa out of the Louvre and build an individual exhibition pavilion nearby for her.

We want to build a new exhibition space which is closely related to Da Vinci and Mona Lisa. By putting all the paintings of Da Vinci’s into one of the AI tools PGGAN and weeks of deep learning, we got 300 2D images, which is what AI learned about Da Vinci’s style. We then took a further study on those images. By comparing the new generated images and Da Vinci’s original paintings we found out a new display of sfumato from AI. It is more abstract but formally similar to the original paintings of Da Vinci’s. We extracted a unique weaving technique from the ‘new sfumato’ and then used this new technique to finish our design of the new exhibition pavilion.

The new pavilion is located on the historical axis of Paris. It was at the original location of the Tuileries Garden which is about a 5-minute walk from Louvre. We hope this new pavilion will become a new hotspot on the axis to activate the surrounding space.