Trained on ~7k AI-generated images of feral ponies. Unfortunately leads to an "AI style" of them, which can be mitigated by lowering the weight.
Can do SFW and NSFW, seems to do NSFW by default
⚠️If you have captioned datasets of feral ponies and want to share, leave a comment, I want to try training this on a better dataset. ⚠️
Description
FAQ
Comments (8)
What's this lora for? /s
I think CivitAI is weird about ponies, archiving this. But very based fren, would have just scaped e621 though, artifacts it learned are really bad. Maybe post to HF too, they're less bad these days, as long as you censor images to death, they don't cry about weights generally too much
Yeah its rough, I used only civitai images since they came pre-captioned, and found an easy way to compile them, hoping someone comes in clutch with a better dataset
@OrdinaryTwig I wouldn't mind trying to throw one together maybe, I have way too much free time to sort though a bunch of it. Are you anywhere other than CivitAI? Which format would you want it in as well? I think most people just use caption txt file + image. I don't have the GPUs sadly because AMD sucks for training, but there is a lot of things I would like so I might send others
@OrdinaryTwig I don't use Discord tbh, I have a personal vendetta against them. Fuck it just use the DMs here I guess, unless you happen to use Signal, but rather not force people to sign up to random crap
used to be a brony but....
Hey, I have a generalized MLP dataset of around 5000 captioned images I made for an MLP anima project I had going on. It includes feral, anthro and humanized images as well as separate groupings for specific characters, scenery, groups as well as a collection of generic anime images for regularization. Each image has a tag caption as well as a natural language caption. Let me know if that interests you.
Sounds cool, my discord is thingfromouterspace1 if you want to provide the files