This week I tried 2 Machine learning tools. I actually tried more but most of them didn't work for me, unfortunately. So I used the one we used in class, The Big Sleep, and Artbreeder. Below are the GIFs of the works that were generated by Machine Learning in The Big Sleep.
"A Russian girl in a pop art style"
"A portrait of transgender man in a pop art style"
"I am just a simple Russian girl
I've got vodka in my blood
So I dance with brown bears
And my soul is torn apart..." - Russian girl song
"A portrait of woman with a disability"
In the Artbreeder I had fun merging Benedict Cumberbatch's photo as Sherlock Holmes and Andrew Scott's photo as Jim Moriarty. These are the parent images I used:
These images are what the Artbreeder came up with after different requirements that I set up for it:
Something I noticed was the bias towards a particular race. When I used the same value for White, Indian, Asian, Middle-Eastern, Black and Latino-Hispanic, it would came up with 3 options of a black person photo that can be seen below. Now I know I might be ignorant on the topic and biased but for me it seems like there is an imbalance in terms of how the algorithm decides on the outcome.
Also the gender biases. When I put maximum value for female and male, the machine had some very specific ideas. A female would be drawn in almost anime style with huge eyes and etcetera, and the male would be shown in a realistic form like a photograph, bald, aged, and with something weird in their mouth. I have no idea what that is but when I reran it, the same thing in the mouth appeared, I just assumed it was a weird mustache. But overall, it was fun, I can create aliens and various creatures with that tool.
Comments