I ran Google’s Deep Dream A.I. on highlights from Donald Drumpf’s presidential candidacy announcement in July 2015, using audio volume to dictate how deep to dream (for each frame). For fun, I used a picture of Cthulhu as a guide image.
### Implementation details
– Source video: http://youtube.com/watch?v=f0UB06v7yLY
– Guide image: http://subgenius.wikia.com/wiki/Cthulhu?file=Cthulhu.jpg
– Google Deep Dream code based on DeepDreamVideo (http://github.com/graphific/DeepDreamVideo) run on AWS EC2 instance w/GPU acceleration
– Iterations, octaves, and nesting based on average audio power per frame (nesting: dreams iteratively run on frame output in multiple levels)