Fellow Canuckistanie welcomen!
LORA's ? what is this?
Thanks!
To create an AI generated image, at least the realistic ones I'm working on and trying to create, you need to start with a checkpoint, think of it as a large base model for generating images, and then you can add other things along the way such as image enlargers etc.
One of the main things you can add into the flow are LORAs. Think of these as smaller sets of trained data with very specific types of output they help guide and generate. Like, there are LORAs that add cinematic grain or lighting effects to the image being generated, or a specific visual style like oil painting or Studio Ghibli or manga. Or maybe you want a Labrador or Malnois as the dog model and not just a generic "dog".
There are a small handful of LORAs to help get dog stuff looking better (I use two of them to experiment with), because the large base models haven't been trained on images like "a dog fucking woman from behind". The large models really have no idea what it looks like physiologically when a woman is being knotted. They often even struggle understanding sexual positions without a specific LORA for "cowgirl" or "reverse cowgirl" added to the flow. And since dog/woman bestiality isn't just run of the mill kind of porn you can see everywhere, makes it even more challenging.
Part of the issue is that there are very few good still images to train these LORAs on as
@allyfitz pointed out in her thread. Even getting a super realistic dog cock is hard because no one has made a model that has looked at 100 different nice photos of canine penises. Would love to see someone like Art of Zoo actually sell LORAs based on their own images.
Not sure what prompts Ally is using to get some of her really good looking images of women being mounted and knotted, but I can generate quite a few in minutes using one checkpoint model so always wind up with 20% that look good.
I know the VFX industry well and sort of got into Stable Diffusion based on discussions with my colleagues. They just have no idea what kind of images I'm making
Hopefully that wasn't too boring!
