CivArchive
    Shanzhagao 山楂糕 / Some sort of style Lora - v1 epoch10
    Preview 458117
    Preview 457929
    Preview 457986
    Preview 457995
    Preview 457918
    Preview 458109
    Preview 458125
    Preview 457919
    Preview 458047
    Preview 457928
    Preview 457920
    Preview 457942
    Preview 457930
    Preview 458011
    Preview 458023
    Preview 458032
    Preview 458208
    Preview 458210
    Preview 458211
    Preview 458212

    为什么叫山楂糕呢?因为它软软甜甜、酸爽不腻~

    Eng description below. 更多例图下面相册请~

    Feel free to merge this lora into lora/models/anything else that I don't know (if only you aren't trying to make money from it of course). 如想与其他lora/模型等混合,只要不盈利——敬请随意。

    为什么上面例图显示inpaint:我用了Tiled Diffusion, 也就是 Multidiffusion 插件来upscale,主要就是会让图片更好看,你可以用常规方式upscale,当然也可以用txt2img产图。

    https://civarchive.com/posts/136081 MD前原图在这儿~

    Reason why the sample images displayed as inpainting: I'm using the Tiled Diffusion, aka Multidiffusion plugin to upscale, you can of course generate with txt2img——myself did that, and sent the outcome to img2img, then upscaled it lel. https://civarchive.com/posts/136081 Original files generated using txt2img here.

    这个风格lora里有一个软软的画风,触发token我设置成了child,会让眼睛圆圆的,腮帮肉肉的,总之可爱的(

    其余推荐使用token: hanfu, red lips, scenery, east asian architecture, light rays, light particles, sparkle, sliding doors, glowing

    推荐CFG:如果没有Dynamic Thresholding就7,但是我强烈推荐Dynamic Thresholding插件,开20CFG,所有东西都好看一万倍(?

    推荐分辨率:啥·都·行·呀,图里你们也看见了啥分辨率都有,就没不行的(草

    推荐模型:我是在based64mix上测试的,风格最贴近原版,不过AOM3也不是不行,其他的自己试试也好啦。

    I've set the trigger token of the "softer" style in this lora as child, makes the eyes larger, cheeks pinchier, everything cuter!

    Recommended tokens: hanfu, red lips, scenery, east asian architecture, light rays, light particles, sparkle, sliding doors, glowing

    Recommended CFG: 7 without Dynamic Thresholding Plugin, 14-20 with Dynamic Thresholding.

    Recommended resolution: literally anything has been successful on my end, you can see it in the preview pics lel.

    Recommended model: I've tested on based64mix, and the outcome is by far the best, however, AOM3、holokukiv2.2、somethingv2_2 could also produce some nice things, feel free to test on other models!

    Description

    LORA
    SD 1.5

    Details

    Downloads
    16,732
    Platform
    CivitAI
    Platform Status
    Available
    Created
    4/10/2023
    Updated
    1/6/2026
    Deleted
    -
    Trigger Words:
    hanfu
    red lips
    light rays
    east asian architecture

    Files

    shanzhagao128dim-epoch-000010.safetensors