之前的 [CUTE] 系列被civitai下架了，因为“Resources designed to depict minors in sexualized or suggestive clothing, poses, or situations are prohibited under our contentguidelines”。
不过你依然可以在moescape使用并下载以前的 [CUTE] 模型：https://moescape.ai/models/959182f4-a1a6-48ea-a6e5-890873d57a65?index=2&modelversion=d7341fb5-0111-4ec1-b4b2-d8978dbdccfb
现在我上传新的 [CUTE] 系列模型，以V4.0为基础做了一些改进，持续更新，希望大家喜欢。
注意：
该模型擅长可爱画风，容易生成比较年轻的角色，禁止使用该模型在civitai上传【未成年色情】内容！！！
质量提示词：
masterpiece,best quality,amazing quality,very aesthetic,absurdres,newest,
反向提示词：
lowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, censor,blurry, watermark,simple background, transparent background,
CFG:4~6
step:30
建议使用高清修复
以下提示词非常有用：
[pastel colors] [calm colors] [anime screencap] [photorealistic]
My previous [CUTE] series was removed by Civitai due to their content guidelines, which state: “Resources designed to depict minors in sexualized or suggestive clothing, poses, or situations are prohibited.”
However, you can still use and download the previous [CUTE] models on Moescape:
https://moescape.ai/models/959182f4-a1a6-48ea-a6e5-890873d57a65?index=2&modelversion=d7341fb5-0111-4ec1-b4b2-d8978dbdccfb
Now, I am uploading a new [CUTE] series model, which has been improved based on V4.0. I will continue to update it, and I hope everyone likes it.
Note:
This model excels at generating cute art styles and tends to create younger-looking characters. It is strictly prohibited to use this model to upload underage pornographic content on Civitai!!!
Quality Prompts:
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest
Negative Prompts:
lowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, censor, blurry, watermark, simple background, transparent background
CFG: 4~6
Steps: 30
It is recommended to use high-resolution fixing.
The following prompts are very useful:
[pastel colors] [calm colors] [anime screencap] [photorealistic]
Description
new v2.0版本由[ZUKI CUTE]系列最热门的v5.0改进而来（现在老的CUTE系列被civitai官方隐藏），此版本在保证生成可爱人物的基础上，强调更细腻的笔触、更多的细节、明亮的光影和更多的姿势。
希望大家喜欢。
The new v2.0 version is improved from the most popular v5.0 version in the [ZUKI CUTE] series. (Now, the old CUTE series has been hidden by the civitai official.) This version, while ensuring the generation of cute characters, emphasizes finer brushstrokes, more details, bright light and shadow, and more poses.
Hope you like it.
FAQ
Comments (40)
I'm overall positive about new 2.0 but I have a few thoughts so far.
First of all, I think the push for added detail makes the generations feel a bit too AI generated -- the model tries to push a lot of small details and it takes more inpainting to fix what ends up being excessive detail noise. A big one for me so far is the model insists on generating 6 toes pretty consistently when I hit a pose that isn't super standard, either in the first pass or the hires fix pass. I'm seeing this behavior significantly more than in 1.0.
Secondly, I feel a noticeable loss of color between 1.0 and 2.0. in 1.0, everything looks more vibrant and saturated. This may be attributed to the more realistic style of 2.0. I'm sure some of this is just my inexperience with the new model.
I will say that I feel like the style (at least the default style) has lost a little bit of the small and cute anime look that this series sold to me on in exchange for added realism. However I will also say when I apply some of my common artist mixes ( ex: (ashiyu \(ashu-ashu\):0.8), (khyle.:0.8), (possummachine:0.7)) it produces amazing results in the style I like to gen in. I understand part of this may be an attempt to separate this version from the old CUTE series so civit.ai doesn't delete your work again.
All this to say, I'm eager for your continued work on the model. I will continue to spend buzz each time I see a new verson. Lol.
Thank you for your evaluation.
Can't get it to play nice i had it running all evening yesterday. In fact it's so poorly following prompt that with sfw:1.4 in positive prompt and nsfw:1.4 in negative prompt and cfg scale 6-8 it still generates double exposure artifacts of missionary pov when trying to gen sfw activities. I just cannot get it to look clean after generation unless i lean into those overtrained nsfw compositions. that's disappointing ngl.
Thank you for your feedback and for using the model. I’m sorry to hear that you’re disappointed. To be honest, I’m a bit puzzled because I haven’t encountered similar issues during my own usage. It might be that your prompts aren’t specific enough. Perhaps you could refer to the prompts used in the showcase images. If you’re using comfyUI, you can directly drag the image into the interface. Alternatively, try different sampling methods—I usually use Euler a. Thank you again for your comments and patience. I highly value your experience and will continue to work hard to optimize the model.
（The above was translated by AI. If there are any impolite or incorrect parts, I apologize.）
Instead of using "sfw" I would instead use "general" in the positive and "questionable, explicit" in the negatives. Those are the actual rating tags that illustrious models uses for rating based content.
Be aware to not add loli as thanks to danbooru which is the most commonly trained on anime dataset the term loli is used mainly for NSFW images instead try using child assuming you're trying to make a loli character.
@dog66609675 i'm trying to make a mermaid doing a handstand on a skateboard, lol. I feel like this checkpoint is a one trick pony.
@MetaGen I feel most anime models would struggle with something so random without some sort of aid. Skateboarding in general has been one I notice anime models struggle with. Let alone handstands on it. With a mermaid. Lol.
@sansmia I used mermaid character lora + skateboarding lora + even tried img2img with skateboarding picture high denoising strength. The model is just overtrained on some poses which makes me thing it was fed bulk ai generated images as training data.
@MetaGen Well, part of the problem could be Lora compatibility. Ideally, the Loras you use are all trained on the same base.
In terms of the very complex pose you want, I'd recommend putting the denoise strength to something between 30 and 50 to help ensure it keeps the pose, and if it didn't change the image enough, just put the new image in and generate again.
You can normally get any pose you want, even with crudely drawn images using image to image just be sure to add some basic shading so the AI has something to work with otherwise at lower denoise it won't change much and be sure to describe what you can and use danbooru tags if one exists.
Of course you can always just try an older version on the model or use a different one.
@dog66609675 Yeah really if I want to complete that picture I will have to generate a depth map maybe using flux or hidream then use that as controlnet pose, or just use a better checkpoint, which is kind of my point. When I was trying to generate this stuff with half decent prompt syntax and correct lora settings, trying different samplers i could see the checkpoint bias was bleeding into my generation. I think it could use a keyword update and more varied training ^^
Where do we get the tags from? Danbooru?
booru tags works for me
unfortunately, any mildly suggestive image generated with this model automatically get flagged (need manual review)
I want to showcase my character LoRA using this model which looks much better, but I cant. so i got forced to use other inferior model.
If you curious, the character default outfit are sleeveless crop-top turtleneck, which is enough to be categorized as "suggestive content" by civitai.
I noticed the same thing. I created a big breasted, obviously older character, but it has a foot visible and 100% covered cleavage. Still manual review! Civit.ai is a joke sometimes.
@sansmia The funny thing is all r+ contents are removed from v1, whatever topic it is.
I don't really do anything here on Civitai aside from checking new models every now and then, but I just wanted to let you know your models are absolutely stunning!
No other models from what I've tested so far comes close to the quality and detail your models are capable off. Although I've not used the latest version a lot yet, it seems extremely good from the little I have used it so far. Fantastic job and I look forward to see your future models as well.
Wow, I found some contents are hidden by Civit, but I can still open X and XXX filters to check these content and model itself.
what how
@girl_likes_math Their new SoT... really annoying
有個問題問一下大家, 泳衣或連體衣都有高光或反光的問題,有辦法解決嗎?
另外,zuki的模都好棒,都下載一遍來玩.
hey~I have a question : swimsuits and bodysuits often have issues with highlights or reflections — is there any way to fix that?
Also, Zuki's models are amazing! I downloaded all of them to try out.
只是思路,1改天气如阴天或多云避免阳光直射,2加亮度Lora,对比度Lora,曝光度lora,负面加强光和反射,3改衣服材质
@1575796814217 有效呢!thanks! 把"shine,gloss,reflective clothing"加到negetive可以有效解決.
Cute checkpoint
Why this model be hidden? Fxxk new TOS!
Because of new TOS, if you want to see this model you need to turn off X and XXX rates.
Please add backup on hugging face, it's more reliable there in the long run.
Take a look at Civitai's new policies. This place is going to become a nunnery. If the skirt was 3 mm above mid-thigh, you're banned!
这个系列还会继续更新吗
i love this model, u can add to Female Rover and female characters of Wuthering Waves in new v1.0? plz
这个系列很棒，为什么都搜索不到了呢。还会继续更新吗？
几十天没看c站又发电了？连更新通知都没了
Love your work man you should definitely back it up civitai is a fuck
Zuki is free ✧｡(｡•̀ᴗ-)✧｡
Damn, it is hidden again T,t
To view this model, must enter the hidden command
what is it?
我想請問一下comfy:10 Nodes 是要調什麼參數? 我剛剛從Stable diffusion轉過來 不太了解
Excuse me what parameters should I adjust for comfy:10 Nodes?
Why do i have to bid just to use a model? This is stupid, this website is stupid
Great model!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.