In August 2015, Leon Gatys, Alexander Ecker, and Matthias Bethge posted “A Neural Algorithm of Artistic Style” to arXiv. They demonstrated that a deep convolutional network could pull apart the content of an image - the objects and their arrangement - from its style - the colors, textures, and brushwork - and then recombine the content of a photograph with the style of a painting.
The key insight was that the two properties live at different places inside the network. Content is captured by the activations of higher layers, while style is captured by the correlations between feature maps across layers. By starting from noise and optimizing an image to match the content of a photo and the style statistics of, say, a Van Gogh, the method produced a new picture that looked like the photo rendered in the artist’s hand.
Neural style transfer became one of the most visible early demonstrations of generative deep learning. It powered a wave of consumer photo-filter apps and showed, in a way anyone could see, that style could be treated as a measurable, transferable quantity rather than an ineffable human signature.
Why business readers should care: style transfer turned an abstract academic result into a viral consumer feature within months, an early template for how AI research moves from arXiv to app store.