5.0增强了写实风格，原始画风更靠近2.5d，由于个人爱好提升了丝袜质感>_<。
5.0 enhances the realistic style, the original painting style is closer to 2.5D, and the texture of stockings has been improved due to personal hobbies>_<.
4.5增强了细节和颜色表现。
4.5 Enhanced detail and color expression.
4.0增强了对比度，进一步解决崩图问题。
4.0 enhances the contrast ratio and further solves the problem of image collapse.
3.0改善了人物表现能力，解决了一些2.0崩坏问题。
3.0 improved character performance and solved some 2.0 crash issues.
2.0：色彩浓郁增强细节，推荐先用Euler，如果色彩太强烈改用Euler a。
2.0: Rich colors enhance details,It is recommended to use Euler first, and if the color is too strong, use Euler a instead.
可以到这里在线生成
You can go here to generate it online
NAIXL_MMMMix - v6.0 | Stable Diffusion Model - Checkpoint | Tensor.Art
使用NAI-XL的eps版本，1.0,7.5,0.5融合（Use NAI-XL 1.0, 7.5, 0.5 fusions）你可以按照自己的方法生成图片，效果都不错（You can generate the pictures according to your own method, and the results are good）
分辨率（resolution）：832 x 1216（建议放大1.5倍数），1024 x 1536
采样器（Sampler）：Euler a （更清晰一点）或者 Euler（绘画），Steps: 28, CFG scale: 5.5,
正反提示词（Positive and negative prompts）
masterpiece,best quality,amazing quality,very aesthetic,absurdres,newest,
worst quality,low quality,logo,text,watermark,signature,
推荐几种画风标签，（Recommend several painting style tags,）
无标签（No labels）
1
(yoneyama mai:0.5),(WANKE:0.5),(noyu (noyu23386566):0.5),
2
(houkisei:0.5),(umehara sei:0.5),
3
(aos:0.5),(fajyobore:0.5),(chiaroscuro:0.5),
4
(ke-ta:0.5),(yoneyama mai:0.5),
5
(rurudo:0.5),(ask (askzy):0.5),(nekojira:0.5),
6
(kinhasu:0.5),(momoko_(momopoco):0.5),(yoshinari_you:0.5),
7
(yunsang:0.5),(ningen mame:0.5),
8
naga_u,[tyakomes],henreader,baku-p,
10
(by tony taka:0.5),(amashiro natsuki:0.5),
11
1990s \(style\),anime coloring,Yuuki_Mizuho_OVA,
12
(soejima shigenori:0.8),(zombie ke:0.2),
13
{artist:Endou Okito},[artist:as109],[artist:ciloranko],[artist:oyariashito],[[artist:cutesexyrobutts]],
14
(ask (askzy):0.5),(yoneyama mai:0.5),(fuzichoco:0.5),
15
(miwa shirow:0.5),(batta 16-sei:0.5),(otokoe:0.5),
Description
FAQ
Comments (7)
现在画面色彩挺好的。就是人物头发不太好。像是一块块布条披在头上，不像是飞舞的丝线。可能是为了表现头发的浓密吧。不过还是忍不住想到抹布
I'm NAIXL_MMMMix glazer #1,
测试了很多厉害的模型，这个绝对是我目前最喜欢的。模型非常多功能，色彩很漂亮，身体的一致性很好，也能生成很美的风景。做得非常好。
Tested many Illustrious models, This one is hand down my favorite so far. Very versatile model, Beautiful colors, consistency of bodies, can do nice scenery. Very well done.
This is an absolute masterpiece.
Thank you for updating the model. However, Ver 5.0 leans too much toward a 2.5D style, making the default art style more dominant and reducing the accuracy of artist tag reproduction. Personally, I prefer Ver 4.0 or Ver 4.5.
Holy this Model is next level.
Best part is that the background looks good as well
Great model!
One issue though, when doing img2img (specifically on 2048x2048 res image, unclear if that's relevant) I got the old SDXL error "A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."
This was an issue with the original SDXL model where the VAE didn't work in fp16, and if I take the fixed SDXL VAE sdxl_vae_fp16_fix.safetensors and rename it to naixlMmmmix_v45.vae.safetensors to override the included VAE the img2img-generation works as intended.