Pony Pube Slider

Finally giving TI/embedding training a try. It's pretty similar to Lora training.

First up is some kind of Pubic Hair slider. I was inspired by this Lora (Fluffy Pubes), and wanted to try my hand at recreating it in TI form. Training images are based off of outputs produced from that lora, which is why they look a certain way. The author was kind enough to provide his dataset, but I ended up just creating my own. I wanted to reproduce the look from the lora, so I wouldn't say the pubes are "realistic", they're definitely stylized.

As the name suggests, it's got some slider capability. Lower weights add a bit of pubes, higher ones give you the whole bush.

It works fine without tags, but adding "pubic hair" and "happy trail" might augment your outputs a bit. The carpet matches the drapes automatically, so no need to tag colors. Though you can add "mismatched pubes" to break that connection.

I'm still new to training TIs, so there's some weird issues that might pop up when using this one. It's a bit random for me when it happens, sometimes it does it, sometimes it doesn't.

• First one, is can sometimes change the hair similarly to the pubes, in that it makes it feathered and wavy. I'm not entirely sure what causes this. Dataset's hair is pretty normal, so who knows.
  Tossing "wavy hair" and "curly hair" into the negatives seems to nip it in the bud.

• Second one, at higher weights (past 1.5), some "furry features" might crop up for some reason. Cat ears and the like. Again, no furries or catgirls in the dataset, so why that's happening, I've got no clue. As before, tossing the offending tag into the negatives fixes the issue.

This does mean that it doesn't slip in seamlessly into wildcard setups if those issues happen, but considering it's a 64kb embed, I'm willing to give it a pass. I'll see what I can do on my end with the dataset to mitigate the issues I've mentioned.

Do let me know if you've got any other issues, or any other conceptual TI's you'd like to see. I've got a small list of concepts I'd like to train, but if there's enough suggestions I might give some extras a try.