Pix2Pix
I tried to make a dog cat.
It's interesting how tiny gestures can greatly affect the certainty of the resulting image, and what it interprets as negative or positive space. I tried pretty hard to get it to interpret lines as certain things
GANPaint
It took me a little while to figure out that the painting here can only really alter an already existing element so that it becomes something else, and not really create new elements in most cases. It's also interesting how the generated fields always seem more flat than the original image.
ArtBreeder
General:
Portraits:
The set of portraits to crossbreed with are almost all really attractive people. It is kinda hard to make a portrait without beautiful eyes. The genetic metaphor is apt - you can almost guess what parts of each image are the result of what "ancestor."
Infinite Patterns by Alexander Mordvintsev
GPT-2
"No, Shirley!" I cried as she began to climb over the bed and then started into the chair again.
"If it is too much trouble don't say anything," she said, still sounding like Shirley.
"I may not want to," I promised, "but I can't help it, can I?"
"Yes," she said, and I was quite sure she sounded disappointed.
Sitting there, staring at the empty window and the uninviting sunlight, Shirley sighed.
"Why, why, it's so nice out here!" she protested. "This is our house, after all, and our people, our house! I don't know what to do any more! I feel so empty like this."
"No," I insisted. "This is not our house... it's not my house."
Google AI Experiments
It rarely correctly guesses what you're drawing. You start to notice after a little while what the typical graphical representation of the thing being drawn seems to be. Sometimes it's not the image I have in my head of the thing.
https://magenta.tensorflow.org/assets/sketch_rnn_demo/index.html