Type | |
Stats | 654 1,670 |
Reviews | (113) |
Published | Feb 20, 2024 |
Base Model | |
Hash | AutoV2 9325F0C257 |
Not much time at the moment to create stuff and i have a some LoRAs that take a few weeks to prepare, so i decided to go a bit experimental.
Started out as something completely different that had nothing to do with what you see now, wasn't even trying to make a LoRA from it. Mostly none of the words i used in the example images have anything to do with the images it was trained on, but when i saw the dataset i thought i should maybe train it on those words and use them for generating pictures. It somehow reminded of the safety-holograms on money or passports.
The dataset was like 30 images made out of AI created macro and microscopic pictures of opal like, very sharp crystal structures that seemed to float in a milky pearlescent substance in different angels. Up-scaled and cropped in Photoshop to just show a flat texture to keep everything as random as possible. Sounds a bit overly fancy but i can't really describe it in other words... and, if you use AI to create images you get used to weird words. Actually, after writing this, i should've used this prompt to generate the images :D - The dataset was also not generated with such a prompt.
This thing is super finicky and i actually just want to test it here in there. The images have the prompts in them and show that i went all over the place with the weight setting of the LoRA and words, arrangement of words etc. That is pretty much just me not knowing what to do and it's not really that presentable i think.
Can't really say much more. Good luck.
Pfff, had clip skip 2 for some the images... i hate when i forget to set it back.