底模:Pony

主页:https://civitai.com/models/514891?modelVersionId=573838

版本:epoch 19

I would prefer if you don't use this model for commercial use.

Training

STEPS: 2,212

EPOCHS: 19

Usage Tips

CLIP SKIP: 2

STRENGTH: 0.9

Trigger Words

HU_TAO_PONY_XL, BIKER_SHORTS,

1GIRL, HU_TAO_PONY_XL, BIKER_SHORTS, FLOWER ON HAT, TAILCOAT, TWINTAILS, COAT, COATTAIL, RINGS, SHORTS, SHOES, RED EYES, LONG HAIR,

support me by following me in pixiv https://www.pixiv.net/en/users/34413372 and tip buzz to me if you want more lora made by me.

 

I made 2 versions, 43 epoch and 19 epoch, 19 is more flexible and 43 should be better at making the clothes.

 

I used screenshots from the game and tried to make a lora that follow the character design faithfully. This lora should be good at making from behind perspective where the clothes is consistent. The clothes is part of the model, meaning the lora is not flexible enough, but you can go full nude mode. The model have bias for nsfw, but it is possible to get sfw art.

write 3d in the end to get genshin 3d style and to get more consistency, use coattails or tailcoat to get the two coat tails hu tao clothe have. The model should be able to make decent comic. Just write comic in prompt to get comic pages.

 

start from 648x976, 768x768, 648x978. The model is not good at higher than 768 and will make error and better to just use img2img or highres.fix.

 

I suggest using ponyxl v6 (the base model) or AutismMix sdxl.

 

prompting with epoch 43 is bit tricky, while epoch 19 it was easier to prompt for it. 43 still give good result when you able to understand how the model works. use epoch 19 lora if you want flexibility and use 43 epoch lora if you are not happy with the other one.

 

The prompt to get hu tao and her clothes: 1girl, hu_tao_pony_xl, biker_shorts, flower on hat, tailcoat, twintails, coat, coattail, rings, shorts, shoes, red eyes, long hair,

use ghost in prompt to get hu tao ghost and butterfly to get hu tao butterfly.

for negative use this:

score_6, score_5, score_4, grayscale, lowres, bad anatomy, bad hands, text, error, missing fingers, bad fingers, extra fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad face, bad eyes, blurry,

 

remember to add furry in negative, The base model is bias for furry so it will turn any character to furry character and jumpscare you

声明:本站所有文章,如无特殊说明或标注,均为本站原创发布。任何个人或组织,在未征得本站同意时,禁止复制、盗用、采集、发布本站内容到任何网站、书籍等各类媒体平台。如若本站内容侵犯了原著者的合法权益,可联系我们进行处理。0
发表评论 共有条评论