HyperQuark

So the data set this is trained from is made of images that are or feature geometrics, fractals, kaleidoscopes, or non-Euclidian geometry. Think MC Escher. Chaos.

While testing it I found it quite capable of more.

This collection is a ton of images I made using the lora with minimal prompts and most of them <10% overall prompt weight.

just to test out what it did on its own or with minimal guidance.

It produced mostly what I was aiming for. and by minimal I mean half of these examples I think have prompts with <10 words. at least 20% probably i have <10 characters in the prompt. This was 100% intended as a what does this lora do on its own. there is a smattering of examples with 80-100% prompt weight as well.

all of the images in this collection should open prompt.

From testing; I did find this lora has an affinity for butterflies even if they aren't requested. No butterflies were in the training data.