It's good to make such things easily accessible, but would be nice if you linked/credited Google with the implementation behind this project.

You should also make it clear that processed images are publicly visible (if, indeed, they are).

I'm not sure the owners of the site know I posted it here. I'm not an owner. You should let them know directly if you feel strongly.

For those wanting to create Deep Dream images without the setup hassle, I had a good experience using the Docker image from https://github.com/ryankennedyio/deep-dream-generator

Gives you the IPython notebook so you can install different training sets, try different parameters, etc.

I want to use a different training set. That is the hard part.

Right -- I've had enough dogs and eyeballs.

This is the list. http://image-net.org/challenges/LSVRC/2014/browse-synsets

You can see it had a load of dog specific image sets added to it (Synsets new in ILSVRC2012). What I don't know is whether it's possible to remove all the dog images and remain with the more varied image sets.

It is so amazingly disturbing

https://dreamscopeapp.com/i/ZTBsNTKfTv

Ha, self transforming machine elves filter! DMT fascinates me, and that is indeed the vibe I get from these images.

Another easy way to play with this is the DeepDreamer mac app (http://realmacsoftware.com/deepdreamer/), although it has some size limitations unless you purchase it ($15).