CivArchive
    ChineseDollLikeness V2.0&V3.0 - ChineseDollLikenessV3.0
    Preview 904106Preview 896232Preview 895710Preview 895208Preview 895205Preview 896593Preview 896557Preview 895198Preview 895714Preview 895711

    Instruction of V3.0

    ***The weighting comparison chart is above, choose one you like

    In this update, I have reselected the training imgs, enriched the number of poses, and the girls look very different from V2.0

    V3.0 can be used in the same way as V2.0

    Clip Skip 3 is a good choice as well

    V3.0更新说明

    ***权重比较图在上面,选一个喜欢的用就行

    本次升级丰富了默认姿势数量,并且生成的人物也和V2.0不一样,更加具有东方韵味。V2.0怎么说呢,鉴于现在网红盛行,所以它更像是迎合主流审美的一个版本,并且国内网红的妆容倾向于韩风,所以V2.0和KoreanDollLikeness有点同质化,反正这两个版本挑自己喜欢的下载吧

    除此之外,使用方法和V2.0相同,请参考V2.0的说明

    可以搭配我的两个中国服装lora:旗袍裙 、马面裙,效果还是不错的

    点击下载旗袍裙

    点击下载马面裙

    可以提高CLIP SKIP,但不建议低于2

    —————————————————————————————

    Instruction of V2.0

    V2.0 is almost totally different from V1.0

    This is a lora which can generate beautiful chinese girls/women photos

    According to my test:

    lora's strenth: 0.35~0.7 (0.4~0.6 is the best I think)

    Better use Hires.fix to generate imgs:

    width and height: 512*768

    Denoising strength: 0.45

    Upscaler: SwinIR 4x

    Upscale by 1.8

    Recommend setting the strenth of outdoor environment prompts (e.g. outdoors, cityscape, etc.) higher as well (above 1) when you give the lora a high strenth.

    Welcome to download another "likeness" model of mine—Euro-AmericanDollLikeness

    V2.0介绍

    效果与V1.0有很大差异

    根据我的测试:

    lora权重:0.35~0.7(推荐0.4~0.6)

    一定要用高分重绘生成图片,参考设置如下:

    分辨率: 512*768

    重绘幅度: 0.45

    放大算法: SwinIR 4x

    放大倍数:1.8

    采样算法应该都行

    当你给Lora一个较高的权重时,建议将户外环境prompt(例如outdoors、cityscape等)同步调高(高于1)

    我还上传了一个欧美真人模型,如果喜欢可以下载—Euro-AmericanDollLikeness

    Description

    LORA
    SD 1.5

    Details

    Downloads
    4,257
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    5/24/2023
    Updated
    8/7/2025
    Deleted
    8/7/2025

    Files

    ChineseDollLikenessV3.0.safetensors