This was trained on digital collages made by Pinterest users. The effects can be quite different depending on the weight, as can be seen in the showcase images.
Description
This version was trained on a different and larger dataset than v1, using the same method as my Concretion of Works by Greg Ellison. It's rather unstable, but different effects can be achieved by putting the trigger in either the positive or negative prompt, or excluding it completely. Have fun!