Femboy Chastity

Version 3.0 release tldr

• Updated to focus more on flat/metal cages

• USE CLIP SKIP=2

• Inpainting works with 'closeup view' trigger word

This LoRA is intended to add chastity cages to characters you generate. Is is still a tad overfit so I have added grid images for you to see how the lora affects the base model at varying weights and CFG scales. Some of the model are very hard to prompt (for example deliberate does not like generating cages on the woman class) for that reason you might find that this model is better served as a inpainting lora.

Due to the training images, this model also has a propensity for cutting the heads off of characters, so you may need to use specific prompts to prevent this (i.e. "cute face", "whole body", etc.)

How To Use

Optimal weights for these models are 0.6-1.0 as a general rule. Generally you should use at least the triggers "chastity_cage" and "testicles" together. You might also consider adding keywords like "pussy" and "vagina" to the negative prompts as most of the training prompts consisted of the 'woman' class.

Most of the training prompts for this model were of the template "A woman in lingerie in a chastity_cage, testicles", for that reason it will probably conflict with any LoRAs of female characters.

You can also use this model for Inpainting, just add the 'closeup view' keyword to the prompt, for example the prompt "a closeup view of a chastity_cage, testicles" is usable for inpainting a chastity cage on another image. The same guidelines about the lower weights and CFG scales above applies.

Finally, if you find interesting ways to get consistent and reliable results, I would love to hear about it to improve the model.

Downloads

Please select the version that corresponds to the model you would like to use, currently I've only trained on PFG111, Comix v4.0, Protogen 2.2 Anime, MeinaMix, Deliberate 2.0, and UberRealisticPornMerge. There is also a version trained on the base pruned SD 1.5 model, that LoRA is a good option if you want to preserve as much of the base model as possible (See the gamer girl example in the Grapefruit version gallery as an example of the difference between the Model-specific LoRA and the general LoRA).

If in doubt grab a couple and run some grids to see which works best for you. If you have suggestions for other models please let me know.

Looking forward to seeing what people create <3