StaC's mom has got it going on. I kid.StaC - Standing CunnilingusSince I did a KLP, there must be a StaC LoRA, too. In theory this model should solve some cunning linguistics that would otherwise be lost to translation.In addition, this LoRA might also function as a character LoRA, for Camellya, Yinlin, Female Rover, Male Rover, Ubel, Umeko and more. This was a test to see if I could make a pose LoRA behave like a character LoRA but with lowered Network Dimensions and Rank compared to KLP.If you want trigger words, look at the list, because I'll say it once, lolVersion un: (Translator's note, 'un' is French for 'one'.)This is the base version, and should work just as fine as version un nilVersion un nil: (T-translator's note, 'nil' is NOT French, but Latin for "zero", effectively making this version "one zero")In an effort from stopping random tacet marks from appearing where they shouldn't, additional images were added to the dataset to try and further teach the model what a tacet mark actually is. While that process was ongoing, additional images of Yinlin were added, as well as images of the Male and Female Rovers, and Fenny Golden (since there were already pictures of Cherno Kegaard, and I felt like adding another Snowbreak character). Theoretically, this is a superior version to un, in that there are more characters that it now recognizes, but testing (on Prefect Pony) has shown that the tacet mark issue has not gone away. To be fair though, Prefect Pony fights LoRAs like crazy.Unlike with KLP, StaC was trained with the network dimension only raised to 64; rank kept at the default of 32. This was to see if it would learn all the concepts of the training just as good; it did. A hypothetical model: StaC_un nil ingus would be trained at the default 32/32, and trained on Prefect Pony as a base, instead of standard Pony. The reasoning being, if it works on Prefect, it might work on anything.Version un nil ingus: (Translator's note: If it wasn't clear to you before, it's just the word "Cunnilingus" broken up into segments, for comedic effect)Went back into the drawing board, and relabelled everything, and added a TON more stuff. Not just characters, but also some... interesting poses that may/may not be related to cunnilingus. This version is the penultimate form, and is superior in all regards. This version, though technically is a pose LoRA, is also a character LoRA with a sizeable roster. Look at the trigger words to see the full roster.Contrary to what I'd mentioned before, this version was not trained on Prefect (didn't have the buzz to do that), and was instead trained on standard Pony, and at 64/64 Network Rank/Dimension. My hopes of keeping the Rank and Dimension that high, was for it to retain the mass amount of details, as far as characters and poses go.The Tacet Mark problem seems to have gone away with this revision, and it should only appear on the appropriate characters (Camellya, Yinlin, etc).Note: With all versions, it is NOT advisable to use all the trigger words at the same time, as you'll likely end up with a huge mess, unless you're trying to do a mass collage with all the characters doing all sorts of things.. The trigger words are to meant to be used selectively, so keep that in mind.

StaC

In an effort to solve the tacet mark/chest flower issue, I tried making it a category, dedicating some extremely close up pictures of them, and while I did that, I added in a couple more characters into the dataset. From what I can tell, this model is exactly the same as version un, minus the fact it now knows who Cherno Kegaard, Fenny Golden, and the Male Rover are; it didn't seem to solve the tacet mark problem. Your mileage may vary, but Hassaku seems to understand when and when not to use tacet marks, for both versions of this LoRA.

Files with SHA256: 12c120583591f8f9e3b02c0f767af0d51ea3124659c2b0a6fdb12dc5e8d90270