Excessive Cum (or Scar) - AnXL and Pony

lora based on animagine xl v3.1 "test version" is not satisfying, I might change the learning rate, and change the file these days.test version: [[:cumpool:A]::B] can correct the strange shadows of cum. The range of A is approximately 0.2 to 0.5, and the range of B is approximately 0.7 to 0.99.My training script seems to be broken. Using the early training scripts that were backed up before for training actually resulted in better results. (pony_excessive1 is trained on backup script with bad learning rate, but I think it is better than pony_excessive2, which has a better lr...) Unfortunately, I accidentally updated the backup script, QAQYou may have noticed my naming of Lora files, especially for the Lora of the Animagine model named Myfetish. So the dataset not only contains images of excessive cum, but also images of wounds (bruise/injury/ryona/guro). If you want to try those contents, please try "excessive+ bruise/injury...". Some of the versions also support "pee, pee on body" etc.v1.1Tips: v1.0 and v1.1 could be used together, example: lora:an31_excessive_cum:0.3, lora:excessive_cum_v11:0.9.Reduce cum on background: please use [:(excessive cum:1.1):0.3] instead of (excessive cum).I used the Lora I uploaded last time and some other Lora(s) that were previously trained to do some random merges. In short, the new Lora seems to have reduced its impact on the art style and will not produce bad images when using "dirty" and "cum" in the positive prompt words at the same time.Lower CFG Scale ≈ more cum (but worse hands)"excessive bruise" "excessive injury" "excessive scar" are still available.Problems:"gokkun" doesn't generate cum drinking. It generates more cum on body.v1.1 is bad at drawing pee on body, but you could try it. If you are using Animagine XL V3.1, you could try artist names. I didn't trained with artists names in tags, but it doesn't work bad if I add some artist names in prompt.The blue hair one used "mana (remana)" and "liduke".v1.1(END)Maybe not excessive enough. The main purpose of this model is not actually to generate massive cum..Tips: Animagine V3.1 itself can generate cum bath well at CFG scale 3.5 without any lora. Please don't use "dirty" in prompt when you are generating cum using this lora. The breasts become too large if you don't add "huge breasts" in neg prompt.Problems:Often generate dirty cum(instead of "white" cum)lying: often generate bad images.