[Premiere] Graffiti Styles Get Remixed by Artificial Intelligence in This Psychedelic Music Video

Like AWorkstation.com on Facebook:

For countless graffitos, New York City has been a gigantic canvas tagged in many styles and colors, and this cultural history gets a neural network remix in the new music video for hip-hop artist Krussia‘s track, “Reshape.” Originally from Russia and now based in New York City, Krussia came across the neural network experiments of artist and programmer Gene Kogan a few years ago, and was intrigued by how he had imported various artistic styles into videos, as he did when transposing Pablo Picasso’s Blue, African and Cubist periods onto famous footage of the Spanish artist painting on glass. Wanting something similar for “Reshape,” Krussia enlisted Kogan and director Dan Melamid to create a moving, psychedelic homage to graffiti using the style transfer technique, which is something like Drew Geraci’s work on the neural timelapse of China.

In the video, premiering today on Creators, Krussia raps as he walks through the streets of New York. This is a pretty typical trope in hip-hop videos; but Kogan and Melamid, using a great number of images, map a wide variety of graffiti works onto both Krussia and the streets. Through this combination of technology and traditional painted media, New York seems to breathe with graffiti history.

“I was following Gene’s work for a while,” Krussia tells Creators. “After seeing his experiments with the so-called ‘style transfer,’ the idea of a music video came right away and has floated in my head since. While writing the song as a subliminal homage to NYC, I would have a vision of a shifting fluid or a dynamic projection sort of and Gene’s work made a direct connection to that.”

“After sharing the concept with Dan, who is a longtime collaborator, we tossed some ideas and he suggested graffiti as a connecting theme,” he adds. “We wanted to utilize images demonstrating a range of techniques and approaches throughout time applied on different surfaces—walls, trains, blackbooks. And we also felt that this style transfer process had similar characteristics as graffiti, recreating an existing surface.”

As Kogan explains on his website, the style transfer works are mostly created using programmer Justin Johnson’s code (inspired by a paper by Gatys, Ecker, and Bethge) that “[demonstrates] a method for restyling images using convolutional neural networks.” As he learned from Kyle McDonald, the best images for style transfer are heavy on strong textural components and patterns. In keeping with the open-source culture of neural network art, Kogan has posted instructions for style transfer technique.

“I sourced the vast databases of what is out there online, trying to get a mixture from classic subway pieces and detailed blackbook images, to new school burners, throwups, and tags on dirty walls,” Melamid tells Creators. “I tried to incorporate the work of legendary graffiti artists that have shaped the culture.

“The problem with this technique is that unless you are using imagery that is heavily ingrained in our minds such as Van Gogh’s Starry Night or highly recognizable pop paintings from artists such as Roy Lichtenstein, you lose all context to the art you are utilizing,” he adds. “I came up with a way to show pieces of the original work, but present it not just in a side-by-side scientific way, but something that is more artistic.”

“My opinion might be biased, but I feel visually it challenges the attention of a viewer in a different way,” says Krussia of the music video. “Even after rewatching it multiple times, you still find things that do not change as you would expect them to.”

Click here to see more of Gene Kogan’s work, and here to see more of Dan Melamid‘s work.

Related:

A Computer Turns Home Footage into Animated Oil Paintings

NYC Looks Even More Gorgeous as an Expressionist Painting

China Becomes a Living Painting in This Neural Network Timelapse

Dive into Deep Dream Infinity in These Trippy Music Videos [Premiere]

Source: vice.com

Rating: 5.0/5. From 5 votes.
Please wait...