Pen-Gesture

Hi there, this model is intended to help when generating images of a tag that was understained on illustriousXL ("penetration_gesture").

There was a recent Lora uploaded by another user which was intended to do the same thing, but personally i found it a bit too unstable so i trained one myself, and now here we are.

Before discussing the versions, i'd like to summarize the lora in few points.

• Does it works? Yes, it does

• Is it stable? Yes, it requires minimal tweaks to achieve good results

• Why should i use this and not other ones? At the moment of writing, there's just another lora (for illustriousXL) that tries to achieve this gesture, but in my umble opinion this is more stable and more versatile

• How can i achieve results like the ones in the preview and in the version section? You can use a prompt/settings similar to the suggested ones

• I like the model, how can i help? If you like my work, that required way more than i am willing to admit, just leave a review, upload one of your creation and share it with your friends.

The majority of usage tips are available in the version section, but if you are curious about more tricks, tips and info you can read the section below this one.

Pen-Gesture (v1)

This version is intended to help in the generation of images that use the tag "penetration_gesture", sadly illustriousXL has not a great understanding of that tag, and often produce images of other gestures (the "kitsune no mado" one is pretty common) or even worse it might separate the two words generating an NSFW artwork.

The v1 is able to fix this issue, and even toh it didn't turn out of i had envisioned it, it works pretty good.

The original plan was to sobstitute the tag "penetration_gesture" with 2 other tags "lefthandp3n" and "righthandp3n", so that we could have an even more controlled output.

This plan failed soon enough, because the dataset of merely 80 images wasn't enough to train 2 separate keywords for two concepts so near, but we achieved our objective, the gesture is now fully understood by the lora.

Not using the lora vs Using the lora.

As you can see the lora effectively does its work (and overall is pretty much plug and play) but it might require few tweaks here and there (as everything hand related).

Most of the test were done using Hassaku (Illustrious) and some using WAI-NSFW-illustrious-SDXL.

The settings that i often used were:

Sampler: Euler A
Steps: ~28 ~32
CFG: ~6 ~6.4

The images were upscaled and treated with adetailer just to achieve an even better final result.

Upscaler: YandereNeoXL_200k
steps:15
denoising:0.3

Adetailer:
face
fulleyesdetection

Note: adetailer was NOT USED to help hands in anyway <- So the gesture results are 100% fruit of the lora.

The prompt structure i followed was:

{Your prompt}, righthandp3n, lefthandp3n <lora:Handp3n:1>

Yes, you need to use both of the keywords.

So how do you use this:

activation keywords: righthandp3n, lefthandp3n
lora weight: ~1 (1 is the optimal value)

i often added: "(fewer digits, extra digits,  4 fingers, 6 fingers)" in the negative prompt (tbh it was something that was still there since i tried the other guy lora, but not knowing if it is strictly needed, i am mentioning it)

Note: the output might vary a lot depending on cfg and steps, so if you find an image that you like, but that sadly produced a 6 fingers image, try to tweak those settings a bit, and you will get the desired output, in one or two tries.

Generally speaking i often stay at 28 steps and 6 cfg, and then try the following:

first i try the same seeds with a step of 30. Then if the image is still not what i desire, i increase the cfg to 6.2 or 6.4 <- most of the times this is enough.

Did i test it enough? Well, let's say that my gpu was an excellent choise to use as frying pan xd.

The first test was run at a resolution of 832*1216 and it was the image you've seen before.

Note: the only things between the two proposed images is the presence of the lora and the two activation keyword.

The second test (896*1152):

The third test (768*1344):

The fourth test (1024*1024):

The fifth test (1216*832):

The sixth test (1152*896):

The seventh test (1344*768):

The last and most important test was the one related to known characters, and to test it i've run few generations of images using characters from fairy tails (here's a brief summary, for practicity i've just used the same prompt changing just the character):

Few more notes:

• When using landscape resolutions, keep in mind that the model was not trained using wide images, so it might struggle with those (to help him be sure to use tags such as "multiple_views" in the negative, to avoid the generation of images like the "before/after" ones).

When prompting keep in mind that even toh the dataset was pretty balanced, there were no "dark blue hair" images, so (some times, depending on the seed and the prompt) it might happen that, as in the 1152*896 test, the hair might come out "light blue"

Thank you so much for reading all of this (i know i talk too much, and i often give out too many unneeded details); hope you enjoy this lora, and to see you for my next models.