CivArchive
    沁彩 Colorwater - v4
    NSFW
    Preview 224275
    Preview 224276
    Preview 224274
    Preview 224272
    Preview 224501
    Preview 224271
    Preview 224270
    Preview 224315
    Preview 224269
    Preview 224314
    Preview 224500
    Preview 224499
    Preview 224268
    Preview 224313
    Preview 224312
    Preview 224311

    沁彩

    润物以水,纳彩于心。

    [ENG Instructions below]

    tusi.art/u/607323927751959196

    • 请注意,每个版本的倾向各不相同,如果你喜欢最有透明感的水彩效果,请选择v1,但它适配的模型相对较少。

    • v4是目前最推荐的版本了。

    • 其他版本各有进化的方向,有兴趣欢迎多多尝试!

    • Please note that each version has a different tendency.

    • If you like the most transparent watercolor effect, please choose the original v1, but it fits relatively few models.

    • v4 is the most recommended version right now.

    • Other versions have their own evolutionary direction, so feel free to try more if you're interested!

    • 这是一个水彩风格的LoRA,还在不断完善中,欢迎各种feedback!

    • This is a watercolor style LoRA, still in the process of improvement, welcome all kinds of feedbacks!

    v4 使用说明:

    • v4 版本在适配多个模型的能力上有了极大改进,不再赘述,请参考sample图片就会马上明白了。同样的,也可以和大量的其他优秀lora一起工作,只需要注意组合权重。

    • 推荐设置:权重=0.8-1,CFG=3-6

    • The v4 version is a great improvement in the ability to adapt multiple models, so without further ado, please refer to the sample image and you will understand immediately. Likewise, it can work with a large number of other lora, just be careful with the combination weights.

    • Recommended settings: weight=0.8-1, CFG=3-6

    v3 使用说明:

    全面强化!

    • 通用的模型推荐配置:CFG<7,权重0.65-0.75,CFG与色彩浓淡完全挂钩,但请注意,根据模型不同,有些在7左右就会出现过拟合的现象。但有一些三次元模型会在7之上有比较复古的重水彩表现。

    • 和其他LoRA配合使用时请注意避免加权和过高的情况,一般来说,加权总值不超过1.2比较合适,如果需要水彩风格较多的显现,则需要保持沁彩的权重为最高的那个。

      举例:下图为使用墨心和疏可走马共3个LORA的效果,权重比为0.5+0.35+0.35

    Instructions for v3

    Power Enhanced!

    Generic model recommended configuration: CFG<7, weight 0.65-0.75, CFG is fully linked to color intensity, but please note that depending on the model, some overfitting will occur at around 7.But there are some realistic models will have a more vintage heavy watercolor performance on top of 7.

    When using with other LoRA, please pay attention to avoid weighting and too high, generally speaking, the total weighted value is not more than 1.2 is more appropriate, if you need more watercolor style to appear, you need to keep the weight of the QIN color as the highest one.

    Example: The image upstairs shows the effect of using Coloerwater, Moxin and Shukezouma total of 3 LORA, the weighting ratio is 0.5 + 0.35 + 0.35

    v2 使用说明:

    现在的沁彩进行了各种强化,有了一定的写实能力!

    • 二次元类模型推荐配置:CFG<6,权重0.85-1

    • 三次元类模型推荐配置:CFG<5,权重0.5-0.7

    目前对于人物面部的风格还是最大限度保留了模型的原始风格,v3版会对这部分尝试进行改进,追求写实风格,到时候会将二、三次元分开版本供大家尝试。

    感谢各位的好评!

    Instructions for v2 version

    Now Colorwater has undergone various enhancements and has a try on realistic style!

    Recommended configuration

    • Anime-like models: CFG<6, weight 0.85-1

    • Realistic-like models: CFG<5, weight 0.5-0.7

    At present, the style of the character's face is still the maximum to retain the original style of the model, v3 version will try to improve this part, the pursuit of realistic style.If things go well, there will be two diff versions by the time they are completed for everyone to try.

    Thank you all for your kind comments!

    v1 使用说明:

    • 水彩的风格会根据CFG的值变化,建议体验从2~6的不同变化,更高值的使用请注意是否能与你的模型匹配。总的来说,CFG的高低和水彩的浓淡正相关。

    • 权重推荐使用0.85-1,如与其他lora搭配请自行判断。

    • 由于一些原因,数据集特意排除了人脸相关的数据,所以大部分主模型在人物面部的表现都还是自己原本的特色,请根据需要自己调整即可。

    Instructions for v1 version

    • The style of watercolor will change according to the value of CFG, it is recommended to experience different variations from 2~6. For higher values, please pay attention to whether it can match your model. In general, the high CFG is positively related to the intensity of the watercolor.

    • The weights are recommended to use 0.85-1, please use your own judgment if matching with other lora.

    • For some reasons, the dataset deliberately excludes face-related data, so most of the main models in the character face are still their own original features, please just adjust yourself as needed.

    Description

    v4 版本在适配多个模型的能力上有了极大改进,请参考sample图片。

    推荐设置:权重=0.8-1,CFG=3-6

    The v4 version has greatly improved the ability to adapt to multiple models, please refer to the sample image.

    Recommended settings: weight=0.8-1, CFG=3-6

    FAQ

    Comments (33)

    I_come_for_pussyMar 13, 2023
    CivitAI

    楼主的v4版本是不是弄错了啊,试了很多模型都不生效,chilloumax什么的都不行,v3版本倒是正常

    AlchemistW
    Author
    Mar 14, 2023

    您好,图4的中年男性头像就是使用chilloutmix,请参考一下参数,该图的权重大约为0.85

    bash99627Mar 16, 2023

    v4和真人照片系混起来似乎容易脸上颜色有点诡异,DreamSharp、Protogen、Realistic Vision等

    qikaixuan1987946Jun 23, 2023

    我也是 点上去直接显示的是<try:1>

    U2U2Mar 15, 2023
    CivitAI

    请问图一的女孩子用的是那个模型?

    RedonkatonkkMar 17, 2023

    图一用的应该是Grapefruit (hentai model)

    AlchemistW
    Author
    Mar 17, 2023· 3 reactions

    您好,这张使用的是anything-v4.5-pruned-fp16

    flytoccMar 17, 2023

    我画出来了,但颜色特别淡,不知道为什么

    AlchemistW
    Author
    Mar 18, 2023· 2 reactions

    @flytocc 请检查一下VAE的设定

    flytoccMar 18, 2023

    @AlchemistW 非常感谢!一开始用的SD自带的VAE,颜色很淡。换用vae-ft-mse-840000-ema-pruned[41894976]后,得倒的效果和图一基本一致。

    zcc6534864Apr 6, 2023
    CivitAI

    你好,请问,如果我想要上手并学习怎么利用这个diffusion绘图,应该利用什么资源学习?

    georgeli2023Apr 8, 2023

    互联网资源,学webUI相关看github项目里说明文档,了解模型看每个模型的文档

    IrisIApr 11, 2023
    CivitAI

    很惊艳的模型呀!大大厉害! 我也在研究国风水彩模型,方便交流一下嘛?

    CHRACEApr 28, 2023
    CivitAI

    请问图7/8的底模用的是什么呀?

    lunarharpMay 2, 2023
    CivitAI

    大爱这种水彩风格!

    design_AIMay 4, 2023
    CivitAI

    可以用图生图,照片转这种水彩吗?

    gaoandwen236May 12, 2023
    CivitAI

    V4好像用于图生图有问题

    qikaixuan1987946Jun 23, 2023

    我也是

    imshadownomanJun 28, 2023

    lora调用命令为: <lora:try2:1>

    SimonSnowMay 15, 2023· 4 reactions
    CivitAI

    您好,我是一名独立游戏设计师,我想制作一款港台武侠风的游戏,请问大佬可以帮忙制作一个武侠风格的模型吗?风格近于河洛工作室的《侠客风云传前传》、《河洛群侠传》、漫画《镖人》。如果可以,待游戏完成后,我将游戏与相关美术设定集一并发给您。

    mhoulz19954403805Jun 19, 2023

    我觉得你可以把人物和风景单独训练成两个LoRa,不就可以解决了

    wangyitongruby584Sep 15, 2023

    网上不是有挺多付费训练lora的嘛也不贵 淘宝和国外网站都有

    keepupcoder215May 25, 2023
    CivitAI

    推荐用哪个基础模型呢

    hiishailey567Jun 12, 2023· 1 reaction
    CivitAI

    大佬方便在tusiart也放下自己的模型嘛?最近sd设备和环境搞不下来,打算在在线生成网站上试下。

    AlchemistW
    Author
    Jun 26, 2023

    已经入驻啦,会慢慢都搬上去

    liudhzhyym807Jun 22, 2023
    CivitAI

    画出的图,人物总是侧面待着,加上了looking at viewer也不好使,是模型训练时都是用的侧面图训练的吗,我的描述词是(masterpiece),(extremely intricate:1.2)>,close shot,(upper body:1.2),1girl,long hair,beautiful face,gentle,

    long dress,

    snow,wind,cloud,

    cinematic light,exquisitely detailed skin,best shadow,delicate,RAW,<lora:Colorwater_v4:0.8>,(facing the viewer:1.5),looking at viewer,

    AlchemistW
    Author
    Jun 26, 2023

    请看其他返图,并没有这个问题,可以尝试更换一下主模型,以及改成portrait

    gpazziniJul 20, 2023
    CivitAI

    大佬 能不能来个 橙光古早立绘那种风格的

    brokoliesAug 30, 2023
    CivitAI

    The name should be "WaterColor" not "ColorWater"

    wktraNov 29, 2023

    From my experience, many don't care. Their models are meant for "their" community, not the community at large.

    martinwalsh600314Jun 21, 2024

    Why?

    keyepochJul 3, 2024
    CivitAI

    cool

    voodoopearAug 3, 2025
    CivitAI

    太棒力

    LORA
    SD 1.5

    Details

    Downloads
    39,798
    Platform
    CivitAI
    Platform Status
    Available
    Created
    3/10/2023
    Updated
    5/14/2026
    Deleted
    -

    Files