Neural Style transfer

is this kind of like Domain Adaptation?

I came upon this google deepdream thing.

They train a traditional CNN, such as say GoogleNet. Then remove the loss function, and try to modify the input image to maximize the activation of the nodes. https://github.com/google/deepdream/blob/master/dream.ipynb

So depending on which layers you care about, you can generate some really freaky things.

Keras: https://keras.io/examples/generative/deep_dream/