This LoRa was made exclusively for the https://civarchive.com/articles/2755 contest.

Mix this LoRa with another in this contest >>> https://civarchive.com/collections/91051 - Deadline ends Nov 10 2023

EDIT: Rejected twice, now approved. Still dont know why it was rejected. Enjoy it folks.

119 training pics of very old school 1600s-1900s occult art drawn mostly in old books, only half the pictures are captioned. I plan to caption all of them in the next version, but that may take a few days. Captioning high quality captions is a huge time sink. The training data consists of a very specific type of intricate and super detailed occult art with a lot going on in the image.

Trained on -

Network dimension 64

Network alpha 32

Its why the LoRa size is so large, 400mb.

I intend to update this in the near future and the next versions LoRa size will be lowered by half or more by using Network alpha 16 and Network dimension 32. I set it way too high to milk as much quality as possible.

WARNING: Due to the training data being old scroll and parchment style artwork, this style does not translate well with realistic art, it will cause realistic stuff to look like a 1970s-1980s picture quality movie. It works well with not realistic like cartoons, drawings, paintings, sketches, and anime, etc.

Usage: Trigger word is "0ccult" spelled with a ZERO, not a letter o. I used a 0/zero to differentiate it from data in the models already trained on occult stuff so my LoRa would have a higher priority.

Use a higher LorA strength if you are not seeing old school parchment style art style.

<lora:0ccult_works_of_Art_SDXL:1.2> gave me more LoRa effect whille <lora:0ccult_works_of_Art_SDXL:1> was barely noticable.

Increase the CFG to 8 to get a stronger prompt effect.