Tips (Pizza Thot) (Illustrious)

This model produces generations of Tips, a character from Pizza Thot by agcats.

It was trained on a bunch of random fanart, so it doesn't follow the original artist's style.

This model is still merely OK at generating her teeth, especially with braces, but it's accurate enough that you can inpaint a few times and get a good result even if you don't like the first result.

Unlike most of my models, this one does not use caption augmentation. This means you MUST use the helper words/captions to get a good result, as they contain much of the training.

The main trigger is: t1ps woman, twintails, glasses, blue eyes

To add her sharp teeth: t1ps teeth (this caption also contributes to her looking more accurate)

To add her braces: braces (along with t1ps teeth)

To add her uniform: pt visor, pt shirt, black pants

The uniform can be itemized, only use the clothing items you want. The model should also be pretty flexible with regards to clothing. You may also want to use red shirt if you aren't getting any shirt with pt shirt but this can also make the sleeves red.

NSFW-specific trained captions: shirt lift, pants pull, penis

Recommended negative captions: text, sketch, grayscale, black and white and possibly comic if you want to avoid multiple panels.

It was also trained on high detail close up shots of her face using close up which was the breakthrough that allowed the teeth to be accurately rendered.

When it comes to generating her teeth during inpainting, I recommend using close up along with the main/teeth caption for better results. It might take a few attempts but you'll get a good result if you keep trying. It's pretty decent even without this though.

I didn't use sharp teeth but I may in future iterations as Illustrious has less of a monster/predator/demonic bias with sharp teeth than Pony does. It seems to help so go ahead and try it out, though be aware it can reduce the braces.

Specific mouth captions like open mouth or clenched teeth can help as well. parted lips can also be good but often just has a flash of white teeth visible.

I actually recommend higher strengths (1.0) for the teeth, because it needed training to learn how to render her teeth and reducing the strength to rely on the base model for support is usually going to work poorly. If you need a lower strength, it might be good to use it for the composition and then fix the teeth at a higher strength.

Although not trained on it, you might want freckles, black hair, or dark hair as captions to keep the model on track.