Edit: Alpha v0.002. you can read the description of the model, but essentially I started editing images to make them proper for training.
Suggested negative:
Comic speech bubbles, Comic panels, Original page, Dall-e, Bad quality, black and white, b&w
_________________________________________
Trained over Pony diffusion on Simpsons comics made by Croc, which are NSFW comics.
This initial release was trained on just a few unedited pages. There was no captions other than "The simpsons comic by Croc." and "Comic cover page." when it was appropriate.
So, this lora is terrible, but it's just the beginning. I intend to work on the dataset and edit the images to make them nice. I publish this initial version because it can still generate something.
Description
Original pages still present and tagged as "original comic page"
In dall-e , I generated character sheet of random "unfamiliar characters from the Simpsons". This is a test to see if it can give character diversity.
Most importantly, I added about 40 images of edited panels from pages in the training set. This means that the images are full screen, often have had their speech bubbles removed. It will help greatly to generate neater images.