This one was an interesting experience. I am A Little Disturbed by the results of playing with it.
This post is going to be long so I’ll put the images below under the cut!
I didn’t choose one specifically out of the images I did but was more interested about how each parameter determined how the images would come out. I’ll be frank, I really hate the images the anime girls were trained with and that prompted me to explore what exactly caused each of them to look that way… I specifically played around with the highest and lowest values of each component (ex. I turned every single element all the way up, I wasn’t that interested in making a pretty image but just a weird image).
This was the first image set: Yuck! They all look the same and I really hate how infantilized they looked.
I played a little bit with the chaos bar to change the way they looked:
They look even more disturbing now to me tbh.
After adding a bit more chaos to the image, I started wondering: What images is this model being trained on for it to look… Like this. ( 🥵??? gross!)
Ok. So let’s put the chaos all the way up.
This is honestly a lot more interesting in looks! I appreciate these, in large part because I don’t see any anime girls!
So what happens if I lower it back down just a notch?
Ok. NVM.
At this point, I wondered to myself what would happen if I just inverted each of the sections this would be trained on:
This is actually really cool! I wonder where exactly the hat the first few images have came from :O. Also slightly demonic looking.
After this I got a little bored with the anime set, so I decided to play around with the human portraits:
The first thing I noticed was that there was a category for Latin Americans, and, of course, as a Latina I could not help myself but play with that.
I mean after all, every single Latin American country is ethnically and racially distinct. Also, not all Latinos are Hispanic and not all Hispanics are Latinos. I have always been really interested in what algorithms think people from Latin America are, as even defining race in Latin America is a bit of a tricky subject since each Latin American country is unique in racial admixture and genetics.
This right here is the average model, okay……..
This is the model where hispanic “genes” are removed:
This is somewhat funny to me since I know a lot of Latin Americans that are this pale, of course, I can’t help but feel like this is a bizarre generalization but in part it might also be that I just don’t understand how this model works.
And when they are boosted to 2:
Anyways, weirdness aside, I used the general model to create some images and this is an awesome project!
