These LoRA are intended to increase the accuracy of specific tentacle concepts since they can be frustrating to prompt for, and it can require a large amount of inpainting to get to the image you have in mind depending on the concept you are wanting to work with. I also just personally don't like how bland the default tentacles look in models, so building a LoRA helps to change the style to give more options. The releases are broken up into multi-concepts (megapacks) and individual concept LoRA. I will update this description on big releases such as the multi-concept V1. For individual concept LoRA please check the version details for notes. It is much easier to keep the descriptions clean that way as I put a lot of notes down from when I am testing.
Planned releases for individual concepts include the following as of now. Note these are not in any particular order and may not be released as listed below. I may separate or combine these during testing.
Tentacles Under Clothes
Milking Tentacles
Tentacles over Eyes & Tentacles in Mouth
Breast Fondling Tentacles (Groping/Squeezing)
Futasub Tentacles
Nipple Tweaking & Injection Tentacles
Grinding Tentacles
Huge Tentacles
Multi Concept V1
This was my attempt at creating a LoRA to more easily trigger certain concepts around tentacles. V1 had too many concepts and a result it does not work that well. I will release more versions in the future which fixes this problem by separating the concepts out. The LoRA still functions in some of the ways I had intended however so I am releasing it as a base to build upon later.
I think part of the problem may have been the way I named the trigger words. I may try another release that attempts to combine all the concepts in the future but only once I know individually the trigger words are working accurately.
The tentacle milkers and tentacles under clothes trigger seem to work the best out of the concepts I had intended. It is far from 100% success rate though. Like usual I am posting comparisons below for V1 to give an idea of how it functions.
Multi Concept V1 Comparisons
Description
Second training attempt at the Milking Tentacles Concept. I trained this after a bit of testing on V1 as I was noticing that the accuracy was low for certain camera angles. I cleaned up the dataset and changed the tagging while adding an overall trigger word. In V1 I had no trigger and had instead used tentacle milkers and added a size before it to try and get the LoRA to be customizable on size. That didn't work quite as well I would have liked so I removed that, added the trigger word "mlkaugment" to get the LoRA to activate. I then tried to set weights to "tentacle milkers" with heavier weights on the larger tentacles and lower on the smaller. This does not seem to work as I'm not noticing a difference when changing the weight on "tentacle milkers".
Overall, accuracy was improved to some extent, but control was sacrificed. With V2 you don't need to use negative prompts as often to differentiate between the tentacles at least. I will be refining this with a V3 later. If I can't reliably get the control I want by V3 I may separate this into its own LoRA apart from the tentacles concept LoRA and do different version for size and style.
Future plans for this include:
add in the ability to reliably prompt the size of the milkers
clean up the output for transparency
control the amount of milk sitting in the milkers
get the "imminent milking" tag to work (this mostly involved finding more images for the dataset, as I have very few right now)
clean up the nipple tweak option. this doesn't work very well in V1 or V2