Just when you thought the neural network fad had fizzled, the New York City skyline transforms into a vibrant expressionist painting, and it all begins anew. Much like Drew Geraci in his Prisma-made timelapse video, in NYC FLOW a new short film, artist and filmmaker Danil Krivoruchko riffs off the deep dream phenomenon, turning ordinary iPhone videos of his city into a mesmerising moving oil painting.
A couple months ago, Krivoruchko stumbled upon a video demo put together by a couple tech guys from the University of Freiburg, which demonstrated the programmers' implementation of a network image generator, a neural-style code. In the video, Manuel Ruder, Alexey Dosovitskiy, and Thomas Brox demonstrated how a ‘temporal consistency constraint’ was used in transferring neural artistic style to video files—speeding up the process while maintaining a more consistent image. A month later Krivoruchko noticed that the group had published their project’s source codes on Github, an online repository that basically works as an open source forum for software developers. This source code allowed him to create a base animation layer and start experimenting with stylised filters on a project of his own.
For NYC FLOW, Krivoruchko worked exclusively with slow-motion footage because it supported the code’s “dreamlike effects,” the artist tells The Creators Project. When he finally nailed down a set source reel, Krivoruchko started running tests to determine which style filter worked best with each shot. After he put everything through to render, he worked out some basic color correction and glitch removal to clean up his final product. Now, you can check out the stunning results below:
NYC FLOW is one part of a larger project Krivoruchko started called Deep Slow Flow. The idea is to make the same types of stylised videos with footage taken from around the world. He hopes that new footage will be added daily. Check out the Deep Slow Flow project on Instagram. And for more by Danil Krivoruchko head over to his website.