Use RNN to create or fill in an image
To create a RNN similar to that of the one outlined in the Deepmind paper DRAW (https://arxiv.org/pdf/1502.04623.pdf). It is fascinating to think of a computer generating images pixel by pixel. If this can be extrapolated to generating images ahead say 60 steps then it could also be used to create animations/video.
I am just at the beginning of this process so I am searching for papers and articles/tutorials on the subject.