This sdxl pony Lora model aims to teach the ai what a ponygirl is, as well as trains the ai to understand how a ponygirl can be used. (such as ponygirl riding etc.)
At the moment it has been taught ponygirl riding both with ponygirl on all fours and standing, it should differentiate as well that any clothes specified will apply to rider only and that the ponygirl stays for most part naked.
if tags such as riding, horseback riding, sitting on person, 2girls, femdom etc are used in conjunction with ponygirl the ai should (hopefully) figure out that one girl is a ponygirl being ridden by the other.
if tags such as solo, 1girl etc are used in conjunction to ponygirl then it should know to exclude the rider and to generate an image of ponygirl alone.
if tags such as riding crop are used the ai should understand the rider holds the riding crop not the ponygirl etc.
Lora should be compatible with any model that uses pony as the base model.
I will put further thoughts about the individual models in the model description notes.
Please make sure that you use the keyword tags the model has learned to understand to get the best results. Also tactical use of BREAK 's in the prompt can aid in image generation for all images. at least for automatic1111 which is what I tend to use.
Description
This version is a little rushed, just putting it out there for now, since I know at least some people are eager for it. I used civitai autotagging, which to put bluntly isn't very good (I've trained this on over 200 images I've made and unfortunately I can't upload text documents of the tagging because civit lora training isn't that clever.). At some point I plan to remake the model when I have time with better tagging which will improve things. Even if that means doing copying and pasting all the tags manually into civitai lora training interface...
Also the lowest allowable learning rate on civit ai is an order of magnitude too high considering the number of images I have for training. (idk why civitai has limited the learning rate like this, it makes no sense, I'm assuming it's an accident and whoever put the civitai lora training together doesn't know the ins and outs of learning rates...). So next version might try runpod or something to get around this if I can figure that out, since civitai lora training leaves areas to be desired.
This model is probably overtrained a little as well, due to learning rate limitations mainly. So next version will probably fix this, but for now, just make sure you specify what hair colours you want your characters to have (unless you want the ponygirl to have orange hair. since probably 40% of training images I made had orange hair ponygirls... yeah I probably should make less orange haired ponygirls from now on.) and make sure you specify any details you want or don't want.
It's been trained on base pony, so base pony it'll work best on probably, but my favourite model to use at the moment for it is probably snowpony base model. So if you want your images to be more details and semi (not really) realistic, then I'd recommend trying that model personally.
Have fun creating!