Painting Time Machine

Every Masterpiece has a Date of Creation.

Comment

Inspiration

I’ve always been fascinated by how different eras have a distinct "aesthetic to it". I wanted to see if a AI can also learn to see these historical patterns as well as an art historian could to test the potential of machine learning.

What it does

Painting Time Machine is a website that analyzes the color palettes, textures, and compositions of uploaded artworks to estimate their creation year. By breaking an image down into multiple components, such as edge density and luminance. Based on the components, the AI puts the painting into a time period or era.

How we built it

The AI was built by using the Metropolitan Museum of Art’s API to create a training dataset of hundreds of paintings. I used K-Means clustering for color analysis and SciPy for texture extraction, eventually training a Gradient Boosting model to recognize the shifts in artistic techniques over the last 600 years.

Challenges I ran into

The biggest challenge I ran into was training the model. The model couldn't originally pick up brush strokes and color patterns accurately and place paintings like the Mona Lisa in the 1100s. But I was able to overcome this by integrating high-frequency edge detection and Laplacian variance to evaluate the texture of the paint itself.

Accomplishments that I'm proud of

I’m really proud of reaching a prediction accuracy within a few decades across six centuries of very diverse art. It was also incredibly satisfying to build a UI that feels polished, rather than just a standard data dashboard.

What I learned

This project proved that mathematical properties, like the skewness of a color histogram or the variance of a stroke, can actually evaluate the human concept of style. It taught me that art and data aren't opposites, but two different languages describing the same thing.

What's next for Painting Time Machine

I’d love to expand the dataset to include global art traditions and make it more accurate. I’m also looking into adding Artist Identification to see if the AI can distinguish between a Monet and a Manet purely by the DNA of the edges. I will continue to work with AI and experiment with the many things it is able to do. Hope you guys enjoy the website!

Built With

Share this project:

Updates