CivArchive
    Hadrian Delice SDXL ILL nsfw - v4.0 - v7.0
    NSFW
    Preview 106409804
    Preview 105994947
    Preview 105994922
    Preview 105994942
    Preview 106301764
    Preview 106155717
    Preview 106413460
    Preview 106155718
    Preview 106155715
    Preview 106314087
    Preview 106377868
    Preview 106302482
    Preview 106127250
    Preview 105994939
    Preview 105994948
    Preview 105996727

    V10.1:

    调整稳定度和色彩

    V10.0:

    增加作者:

    by nishimura_kinu

    本模型比较依赖highres,自行斟酌。

    highres推荐先尝试0.6重绘幅度直接放大到长边2048,可以增加很多细节。崩了再降。

    我常用的设置是euler_a,steps 28,还挺稳定的。

    关于加不加这个"by" 既然有帅哥问了,这里也提一下:

    其实加不加都行,加上还原度更高,但是不加可能看起来更好,会把本模型更多元素带出来,具体就是更清晰的画面,更丰富的颜色,其实就是变相降低了权重。毕竟大部分画师的素材是达不到插画级要求的。barbarian_takamoto是个很好的例子,当然我不是说大佬画的不好...

    独家/微调 画师列表:

    by nishimura_kinu

    delice_realistic

    by tabunnanhun

    by kanae_\(cookie\)

    by ame_ame

    by tatami_to_hinoki

    by tnsk

    by n40a1

    by shiro_\(a923808254\)

    by neki_\(wakiko\)

    dragon's_crown_style

    by jima

    by barbarian_takamoto

    by ama_mitsuki

    by alicecrazy

    by betabeet

    by 1ssakawaguchi

    by ushiki_yoshitaka

    by aleriia_v

    by hungry_clicker

    by mankai_kaika

    P+:

    masterpiece,

    P-:

    (worst quality), (low quality), old, oldest, lowres, censored, extra toes, bad hands, extra limbs, patreon, watermark, signature, artist name

    DeliceSDXL artists - collection posted by Hadrian

    • Euler A or Euler.

    • CFG 3-7.

    • Adjusting the weights of negative prompts like 'white background, monochrome, greyscale' can modulate image contrast and saturation.

    • I usually upscale to 1536-2048 with 0.6 denoising to fix minor issues, then use Tile to upscale to 2048-2560+. If the low-res structure is fine, I directly Tile upscale to 2048+,My Tile parameters: strength 0.3-0.4, denoising 0.6-0.7,end at 0.7. If you only focus on portraits and upper body, slightly reduce the upscaling ratio. HadrianDeliceSDXL Workflow sample | Civitai

    • Euler A 或 Euler

    • cfg 3-7

    • 调整负面提示词的"white background, monochrome, greyscale"权重可以调节画面的对比度和饱和度。

    • 我一般先放大到1536-2048,重绘幅度0.6来修复一些小问题,再用tile约束放大到2048-2560+,当然如果低分图的结构没什么问题也可以直接开tile放大到2048+,个人tile参数强度0.3-0.4,重绘幅度0.6-0.7,如果你只画肖像和上本身,请减小放大倍率 HadrianDeliceSDXL Workflow sample | Civitai


    V9.3:

    • 调整默认风格,自己用来画黄色废料

    V9.2:

    • 出图很稳定

    • 个别风格会发粉发紫,P-加个比如(pink theme:0.8)就行,懒得调了,上传一次不容易

    • 如果做highres的时候画面有些脏,把scheduler换成beta可以改善

    V9.0:

    • 增加作者:

    • by tabunnanhun

    • by kanae_\(cookie\)

    • by ame_ame

    • by tatami_to_hinoki

    • by tnsk

    • by n40a1

    • by shiro_\(a923808254\)

    • by neki_\(wakiko\)

    • 增加一个2.5d风格 delice_realistic

    V8.0:

    • 重做 dragon's_crown_style.。真费劲啊,弄了好久。

    • 加强了背景表现。

    V7.1:

    • 增加画师: by jima,

    • dragon's_crown_style,颜色凑合,笔触不行,还得调 (刚刚发现用错训练集了,诶嘿~)

    v4.0:

    • a lazy version for lazy people who don't want to know how to use this with artist names, also facilitates online creation and preview. If you know your way around, try comparing with 3.0.

      In Version 4.0, I've observed that nearly all users - whether submitting artwork or feedback - exclusively use the default style. While I've refined this default aesthetic. But it is subjective.

    v2.0:

    • Conceptual enhancements have been made to the actions of wielding a katana, such as unsheathing.

    • Minor adjustments have been applied to the style of 'hungry clicker'

    v1.4:

    • Adjust 4 artist styles:

    1. Alicecrazy trigger word: alicecrazy style

    2. Betabeet(Anon) trigger word: betabeet style

    3. Ama Mitsuki (Ten Mitsuki) trigger word: ama_mitsuki_style_2024

    4. Tnsk trigger word: tnsk style

    中年失业在线乞讨 Middle-aged unemployed begging online

    buymeacoffee.com/hadrianwu 支付宝: [email protected] 下方有二维码,谢谢!

    本来我已经不打算在C站发模型了,因为我想保留一次警告次数,至少能在C站发个公告,发个动态,详情见我的文章。

    不知道为什么这群老外只会盯着第一条警告辱骂我,我说了,我可以不发图,而且自从收到第一条警告,我也一直只发模型样图。

    我郁闷的是有人在我的评论区发了P站链接,导致我被第二次警告。2024年还要付连带责任的?原来美国还是封建社会吗,笑死个人。况且我也没在任何管理制度中见到这一条。

    过了几个月我也不想说啥了,还是有不少人找我,我就放出来吧。如果有一天我又莫名其妙的被ban了,来shakker(liblib)或者bilibili找我。

    Description

    FAQ

    Comments (19)

    A9616777Oct 21, 2025
    CivitAI

    是额外训练的画师需要加BY,还是所有的画师TAG都需要加BY?

    Hadrian
    Author
    Oct 21, 2025

    其实加不加都行,加上还原度更高,但是不加可能看起来更好,会把本模型更多元素带出来,具体就是更清晰的画面,更丰富的颜色

    DjojiNov 5, 2025
    CivitAI

    man im kinda sold on this can you tell me more about this model?

    Hadrian
    Author
    Nov 12, 2025

    It was originally a port of the Pony version, but now I'm trying to make it produce more of my preferred styles without using LoRA.

    DjojiNov 12, 2025

    @Hadrian one more thing, did you have a list of all artist or some features in your model? or where i can search them myself?

    DjojiNov 12, 2025

    @Hadrian cuz i need to know that before i go for 8.0 or the 7

    DjojiNov 12, 2025

    @Hadrian sorry to bother again but for activate the style for exemple hungry clicker, did i need to put a hungry clicker, by hungry clicker or hungry clicker style?

    Hadrian
    Author
    Nov 13, 2025

    @Djoji The adjusted style is described in the model notes. Trigger words don’t need to be too strict, though they may cause some variation in results. Feel free to experiment — what matters most is whether the outcome fits your own aesthetic.

    dan1982kieenNov 12, 2025· 2 reactions
    CivitAI

    万分感谢你的辛苦工作,它们是有意义的。

    wiz0806Nov 22, 2025· 4 reactions
    CivitAI

    My favorite model. Either using artist tag or not. Please keep up your great work!

    826531309243Dec 18, 2025
    CivitAI

    大佬,为什么我用这个模型,无论是在webui forge还是comfyui,手的质量都很差。我试着用不同的画师tag,或者不同的负面提示词,都达不到理想的效果。我还试了几个在其他模型效果非常好的负面嵌入,在这个模型里就是无法很好的控制手的质量。尤其在使用hands gesture这个提示词的时候,生成的手就不止是多一根手指这么简单了,会扭曲成一团。

    Hadrian
    Author
    Dec 18, 2025· 1 reaction

    9.0确实有这个问题,8.0和7.1会好很多,9.1在修复这个问题,但是我传不上来…200k/s传了6个小时断了好几次,我打算放弃了

    826531309243Dec 18, 2025

    @Hadrian 大佬辛苦了,8.0好像有点吃分辨率,9.0我1024X1546可以直出,之前的只能960X1280然后再放大

    kainnosgoth1988Jan 16, 2026· 1 reaction
    CivitAI

    does it work well with style LORAs?

    Ghost_NameFeb 3, 2026

    Nope

    noescaMar 15, 2026
    CivitAI

    大佬666,本来想融合的,有心动的感觉,但是感觉是个画师集模型比较依赖画师tag,感觉不是很好融合

    Hadrian
    Author
    Mar 15, 2026

    嗯,我知道我想要什么,所以不是很在意默认风格。想融合调整好权重就行了,可以找个固定风格的模型合(大部分ill的衍生模型都有),或者揉个风格lora进去

    Hadrian
    Author
    Mar 15, 2026

    但是话说回来,如果打算出一个单一风格的模型,也没必要用我这个合了…

    noescaMar 16, 2026

    我不是单一风格模型,你可以去看,但是我的模型现在最大的问题是结构崩坏,我不会微调,想试试用你的模型来修正一下。

    Checkpoint
    Illustrious

    Looks like we don't have an active mirror for this file right now.

    CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.

    Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.

    Details

    Downloads
    61
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    10/17/2025
    Updated
    4/27/2026
    Deleted
    10/18/2025

    Files

    hadrianDeliceSDXLILL_v70.safetensors

    Mirrors