This lorA model is an extension of realistic NSFW poses for Z-Image Turbo, focusing on panties pulled down, which the base model has major issues with, just like fantune models. It either has panties, doesn't have them, or only draws part of them. This extension is also imperfect; it requires a much larger dataset than was used for training.

The dataset currently contains 165 selected images, but I plan to expand it to 800 in the future. After testing this lora, I realized what it needs. But that's for later, as I plan to create a similar lora for anime, also for ZIT, and overall, I have other things to do.

Also, I should point out that I have a pretty good PC, and I generate 1280x896 images in 24-25 seconds with a 9-step BF16 model. Therefore, out of, say, 20 generated images, I can easily discard a third or even half. If generating a single image on Q8 or FP8 takes almost 3 minutes, like it does on my laptop with an i7 7700, 16GB DDR4, and a GTX 1050ti 4GB at 1280x720, then it's probably best not to use lora in its current form.

Currently, the available poses are standing, sitting, and on all fours, front and back, and lying .

Take the lower part of the prompt from the examples I provided.

For example,

"panties down, pussy,

very aesthetic, best quality, ultra-detailed, masterpiece, fhoto cosplay, realISM, photorealism, hdr, She stands on all fours and takes off her panties, panties down, pussy,"

this means she's standing on all fours in front with her panties pulled down.

The image prompts have everything; use them as a basis.

Important!! Use the description "whole body or half of the body" (upper part or lower part) in the case of the lower part, just remove the description of the face, so that the head does not grow out of nowhere.

strengt 0.3 - 0,45

Values ​​higher than this often break the anatomy and affect the color correction (color correction often improves, but I don't recommend values ​​higher than 0.7 of the lorA strength)