Check belowing images about comparison between no-lora, version1, version2, and concepts, and v2 compability amoung models.
Compared to V1, I added general images (images without tentacles) generated with AOM3 model and caption them with "style of AOM3" in V2 training dataset, I hope this may help preventing style influence.
Description
「best quality, absurdres, unity cg, masterpiece」for postive and 「worst quality, lowres, bad anatomy, blurry, extra legs, extra arms, artifacts」for negative prompts. Use belowing words about tentacle contents as you like.
tentacles, tentacle pit, tentacle bound arms, breasts apart, suction tentacles on breasts, tentacles around nipples, milking, projectile lactation, tentacle vaginal, cum overflow, cum string, raining cum, spread legs, m legs, torn clothes, torn legwear, clenched teeth, dilated pupils, empty pupils, fainted, unconscious, half-closed eyes, rolling eyes
FAQ
Comments (7)
感觉楼主用的训练集素材,为了照顾各种触手题材但是画风不统一,而且角度和风格区别很大.导致最后出的lora概念模型和其他模型适配性一般.会干扰原模型.
猜测是模型自己没有很好的区分画风和概念区别.目前tag里面play桥段,光翼战姬系列都有,如果都用光翼系列的CG素材去做反而效果能好一些.
在训练lora模型的时候,如果想让模型记住,应该删掉想相关的描述字段.lora本质是小模型,对于概念训练不能给太多画风差异过大的图片.
确实是一个给不同姿势打上不同trigger words看能不能分别理解的尝试
That's so great!
I've removed some of your images within your description for TOS violation.
Sexual or suggestive images of minors are not allowed.
lol the p3dos are mad
prompts? trigger words?
效果很不錯
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.