This model is trained on Public Domain Datasets and available for both SD1.5 and SD2.1, giving very unique images with each version.

v21-DB2 no captions, pure style model designed for Stacking.
v21-style - trained on dataset sharing the same name, BLIP2 captions
(prompting captions produces close variation on ground truth)

Danger Helpers here:
https://civarchive.com/models/20517/danger-zoo-johnsons-negative-embeddings-helper-collection