In 2018 I began experimenting with deliberately mutating and blending the learned weights inside of neural networks I had previously trained.
I call this technique of latent lobotomy "Neural Glitch". Unlike the classic technique of glitching a single image by manipulating the bytes of a saved file this method causes the damaged model to continuously generate misinterpretations on a textural as well as a semantic level.
PNG, 2048x2048 pixels
Original creation date: August 28th 2018