Turns out when you cycle an image repeatedly through the Gooflator, the distortions converge on a stable state after a few iterations.
This is partly because the templates it was been trained on and is trying to impose are orientation-specific... so if you really want to fuck an image over, you rotate it 90° between each iteration.
This is the way to reveal the CODE DEEP SEVEN technology that lurks in the darkness under the table. But to find the sad pekinese faces in the wall, it helps to invert the colour palette before goofling and then invert again afterwards.
There seem to be a lot of wine goblets lurking, latent, in the morphogenetic furniture of our lounge.
13 comments:
Hmmm. The second morphing sequence looks like a nasty case of thrush.
Rule 34 warns us that Google probably have another neural-network image-processor package that has been trained to recognise and classify porn images by superimposing templates of human turgid bits. What *that* will do to an ambiguous background does not bear thinking about.
Seems that the word "eldritch" belongs in here somewhere.
~
Thank you kindly, I didn't know until now that I needed that.
Does it have no setting other than "acid?"
Why would you want one?
Hey, variety worked for spiders, maybe the google would like it too.
Why would you want one?
I dunno, "mushrooms" or "mescaline" or "I haven't really slept in three months" could also be interesting.
I've heard rumors that there are other versions of this things that weren't trained quite so heavily on pictures of dogs or slugs or whatever. (Dogslugs!) I don't think they are actually public, though, which makes me wonder what exactly they were trained on, of course.
I think tigris is asking for the spider-trained version.
Did somebody say spider?
Jennifer will find that very helpful, OBS.
I am always very helpful to those with seizure disorders and/or arachnophobia.
...an inordinate fondness for armadillos?
Post a Comment