This was made mainly for slime girl penetration. I have not tested other cases where transparency would make sense.

v4 update:

Weights above 0.8 won't cause quality degradation anymore. A weight of 1.0 should work fine for most things. I also decided to manually fix all of the dataset captions and retrain from scratch with a new method. It does better at affecting only what it's supposed to now.

Getting proper transparency to work with slime girl penetration is a pain. Most approaches either use x-ray (which looks weird for slime girls), do some complicated prompting workarounds, or just deal with low annoyingly low success rates.

This is my first attempt at a LoRA so I can't guarantee it will work properly for you. The dataset is a mix of different aspect ratios, slime colors, and positions. There also aren't any oral or double penetration images in the dataset (for now) so those might not work either. I added more penetration types in v2 but they are still WIP. The dataset isn't finished (and this is still true for v4 since transparency with fingering almost never works right).

translucent_penetration should be the only thing needed to trigger it, but you might get better results with additional tags/phrases such as slime girl, transparent skin deep penetration, perspective, monster girl, (color) skin (such as blue skin, green skin, multicolored skin, etc), penis visible through body, etc. Some penetration types might work better with specific angles (especially the types that are still WIP).

I recommend pairing this with a dedicated slime girl LoRA (such as DaSiWa-Illustrious-Slime-Girl, Realistic Slime Girls, or Slime Girls - Concept) and a checkpoint that is able to handle NSFW concepts well such as DaSiWa Illustrious | Anime (SweetPoison was used to train v1 & v2. AlluringAbyss was used to train v3). You might get some noticeable distortions and quality drops with other LoRAs if the strength is too high. Setting it to ~0.7-0.8 is enough to basically guarantee the concept in most cases, but a good LoRA combination with some good prompting can get it to work with as low as ~0.25.

My typical generation args are either:

steps: 37
cfg: 4.0
sampler: euler ancestral
scheduler: simple

If you are having issues with four or six fingers and you are still using v2, try updating to v4 and reinforcing your negative prompt.