AI artists creates hypnotic 'animated watercolours' of New York

The project uses techniques revealed earlier this year by German researchers. To make it work, the researchers could not just apply the algorithm to individual frames, as this would lead to flickering or and rough transitions. So, they introduced a temporal constraint that'penalized' deviation between two frames, meaning it considers the flow of the original video to preserve smooth transitions. In order to improve the method's consistency over larger periods of time, they incorporate long term motion estimates. They also developed a multi-pass algorithm to work around the'artefacts' created during style transfer The video was made as a part of the Deep Slow Flow project, and uses video processing methods revealed earlier this year by a team of German researchers to apply distinct styles to the clips. It begins by panning across a colourful view from the Hudson River, with One World Trade Center visible in the background.