Intro

CLIPig generates images by using the CLIP network as an art critique.

A bunch of pixels is continuously adjusted to increase the similarity of their features with some user-defined target features. Both features are derived via CLIP.

Through backpropagation, the most common method of training artificial neural networks, the dissimilarity of trained features and target features is translated back into pixel values which adjust the initial bunch of pixels just slightly. If we do this long enough, with some artistic variation in the processing pipeline, an actual image emerges.

CLIPig is designed to allow a lot of control over those variations which requires a bit of documentation.

Please browse through the walk-through to get an overview and follow the links to the reference pages and any point.