I could make work this stuff in my python, but it just adds dogs, or dog eyes, and some pagoda's, this gets soon boring
I see you can create a file training it with images, not had time to use it but seems just matter of having hundreds of images to pass to it to learn - big work is to pick the images to pass to it.
There are a few different "learned/trained ai files" already done, you can use instead of the doggie one, this site let select from a list (
WITH THE POWER FROM BEYOND THE GRAVE WE SHALL AVENGE OURSELVES ) , but really this deepdream it will be fun when people will provide files trained with tits, obama faces and other specific stuff that is not dogs. So pics are tittized, obamized and so on. Now we just have this dogs and pagoda thing I think it shows 5% of the possibilities. PS: IT seems that black & white blurred images make best results, colored and sharp images so so.