The videos are generated using a Keyframe that is either painted by hand or have edited with some specific style. Using 3D Neural Style Transfer, it's possible to take the information from 1 frame and apply to the entire animation.
π Published paper about the technique:
https://dcgi.fel.cvut.cz/home/sykorad/Jamriska19-SIG.pdf
π Suplementary Material:
https://dcgi.fel.cvut.cz/home/sykorad/Jamriska19-SIG_supp.pdf
π¨βπ Researched and implemented by Daniel SΓ½kora for the Department of Computer Graphics and Interaction,Faculty of Electrical Engineering, Czech Technical University in Prague and presented on the SIGGRAPH 2019 conference.:
https://dcgi.fel.cvut.cz/home/sykorad/