That is freekin sweet
an addon for blender will be interesting
Really awesome!
My concern is, how much human work needs the system to really modify the original video in order to get the result? If it would be 0, that would be nearly scary 
Not that much scary, actually, seems it still needs a human to paint the reference picture first. 
I have tried this
but it is so hard to get everything working for windows 
Got to say that’s pretty cool. War more impressive than the Google demos earlier. And even found my own shot from Sintel in there! 
Tried it too, but then changed to Ubuntu - its far easier there.
This is a very impressive clip, I am thrilled to see what neural networks will achieve in the next years…
For me, it is also somehow scary 
On the other hand, it opens quite some possibilities. When you create a design concept, you might be able to get a first impression fairly quickly. You might add some generic figures into the scene, maybe even roughly animate them and then show it in the style you want. The result might be good enough that you can already see what works and what doesn’t. You may even adjust the style or try completely different styles fairly quickly.
I believe that this kind of technique will help in the future to improve workflows, especially in the early production stages.