How did AI painting take off? From history to technological breakthroughs, read the history of the hot AI painting in one article

Behind this change, what exactly has happened? Let's take a comprehensive look at the history of AI painting, and then understand the breakthroughs in AI painting technology that have taken place over the past year or so.

The History of AI Painting

AI painting may have emerged earlier than many people think.

Computers appeared in the 1960s, and as recently as the 1970s, an artist, Harold Cohen (a painter and professor at the University of California, San Diego), began building the computer program "AARON" to create paintings. But unlike today's AI paintings that output digital works, AARON actually controls a robotic arm to paint.

Harold's improvements to AARON continued for decades, until he passed away. In the 1980s, AARON "mastered" drawing three-dimensional objects; in the 1990s, AARON was able to paint in multiple colors, and it is claimed that AARON is still creating today.

However, AARON's code is not open source, so the details of its painting are not known, but one can guess that ARRON was simply programmed in a complex way to describe Harold's own understanding of painting -- which is why, after decades of learning iterations, ARRON still produces only the colorful abstraction that is Harold Cohen's Harold Cohen's own style of abstract color painting. Harold has spent decades putting his own understanding and expression of art on the canvas through a program that guides the robot arm.

Although it is hard to say how intelligent AARON is, it is fitting that it was the first program to paint automatically and actually paint on canvas, giving it the title of the originator of AI painting.

In 2006, a computer painting product similar to ARRON appeared, The Painting Fool, which looks at a photo, extracts the color information from the photo, and uses realistic painting materials such as paint, pastels, and pencils to create a painting.

These two examples are the more "classical" way of automatic computer painting, a bit like a toddler, a little bit like a baby, but from the point of view of intelligence it is quite elementary.

Nowadays, the concept of "AI painting" refers more to computer programs based on deep learning models for automatic drawing. The development of this painting method is actually relatively late.

In 2012, two of Google's biggest AI gurus, Enda Wu and Jef Dean, conducted an unprecedented experiment, using 16,000 CPUs to train one of the world's largest deep learning networks to guide computers to draw pictures of cat faces. They used 10 million images of cat faces from YouTube and 16,000 CPUs for three days to produce a model that was so exciting that it could produce a very blurry cat face.

In today's world, the training efficiency and output of this model are not worth mentioning. But for the AI research field at that time, it was a groundbreaking attempt that officially opened up a "new" research direction of AI painting supported by deep learning models.