you trained it? can you share the trained file?

Awesome pics!

Because many people have asked how to make a picture, so here I provide the myself-training pt I used: the prompt after 400,000 trainings, headless-400000.

https://drive.google.com/file/d/1shm5YDFlHQTT9SqxMRHSMRE-SCeyzdJE/view?usp=share_link

Basically, this is an ai training set that replaces the head of the original character with "a ball of blood" or "a bullet hole" or "a burst of blood" as long as prompt "headless" is entered. Its potency is quite good, and it can also be used to refine blood stains if the intensity is reduced.

The pictures I posted this time are all generated by this pt with various clothing tags, and there is no modification other than that. However, when I do formal works, I usually retouch the pictures several times to enhance the effect.

However, this prompt is not a panacea, and it also has many flaws. It is like saying that most of the women in the beheading picture are not wearing clothes, which leads to headless often taking off the women's clothes by the way, or half-dressing them. Furthermore, because most of the reference R18G samples are large-breasted women who were beheaded, so it is not very good at drawing flat-chested Lolita (if you have to do it, please move the order of keywords back, and wait until the characters in the whole picture are formed. beheading), which has the effect of filling the character's chest with water and increasing it. It is also not very good at drawing poses other than the front, and it is easy to collapse due to changing actions.

However, if it is only to produce a headless girl's body, it can do it perfectly. For me, this makes up for the lack of NOVEL AI.