Unlike most of my other abstract models, this one was trained on representational abstractions so that images should appear to be of what you tell them to be of. However, the images I used were of widely different styles, so generations should have unexpected stylistic combinations.Sometimes it may prove useful to include "simple background" in your negative prompt.

Files with SHA256: 13faf085f6c80d48de0bc09f501a08ae10839aa36bc5644c717dd17dc0df8d11