目前使用了二十多个光辉底模的模型，大佬这个是最稳定的

For people who have trouble with Illustration models:

- this is for not only this model, but all Illustration models (this model in v6 btw)

Steps are quire universal and can be used from 20 onward.

Hires.fix works very well and I would recommend it.

CFG scale can be anything from 3 up to 7, so just test out what works the best. If you get artifacts, turn down the CFG scale and try again

I would personally start with 1024x1024, because that is one of the things the Illustration model was trained on.

Sampling works best with either "Euler a" or "DPM++ SDE" from my testings, but "DPM++ 2M" will also work.

Most LORAs work very well, but slightly too well in some cases. I would turn down the strength of LORAs by 0.2 or 0.3 of the value you would use in PONY models.

While not needed, I still recommend ADetailer for the faces.

Common positives are like in SD1 days "masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, " and for negatives "lowres, (bad quality, worst quality:1.2),bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, (censor), blurry, watermark, simple background, transparent background, multiple views,".

Hope this helps some people here!

Thank you for this model!

I have a little question. It seems that the sample images are generated by 'External Generator'.
Which extension is used to generate images?
I've used WebUI interface.

One of the finest illustrious-based models, keep going~

This model is amazing. I've been stuck in the Pony ecosystem for several months at this point and this model was the first time where I felt like I could do what I wanted. Weird, funny, amazing concepts that Pony hadn't the slightest inkling of. The range of options is so far above and beyond. It's the first time in a long time I've been so excited for the possibilities. I hope the Illustrious ecosystem really takes off.

I prefer V5 to V6. I like V5 better than V6 because the background and the transparency of the clothes come out more beautifully. But that is subjective, so I would like to hear other people's opinions.

This is the new G.O.A.T. model. I've been using Pony since its release, but this model is so much better that I'm switching. I hope people will embrace it with the same enthusiasm they had for Pony.

By the way, should we use the SDXL VAE, or does it already have one baked-in?

Personal usage experience:

个人使用心得：

"very aesthetic,absurdres,newest" may significantly affects the style, espacially combining with artist tag. You can try to omit them if the artist style is not strong enough.

The model style is very stable and even does not require negative prompts. Negative prompts may have an impact on the background.

--

"very aesthetic,absurdres,newest" 可能显著的影响画风，特别是和 artist tag 一起用的时候。觉得画风强度不足可以去掉试试。（个人觉得这几个 prompt ，很难感觉提升，但能感觉画风下降，综合坏处大于好处）

模型稳定性非常牛皮，大部分情况可以不用负面词。使用负面词可能会对背景有影响，更容易生成单调的背景。

Amazing job! V5 quickly became one of my favorite models (currently my main model, in fact). It's the first model I enjoyed these past few months that isn't a Pony finetune. The results are often great for both simple and complex prompts.

The only issues I have with V5 atm are the fact it has some trouble with penises (flaccid penises or erect penises with the foreskin not pulled back, for example), and the accuracy when it comes to some less popular characters. But both of those things can easily be fixed with LORAs.

This model works very well at high resolutions. It is quite suitable for Upscale. It also seems generally successful in creating the background.

It's also pretty good in the hands. It works fine as long as the hands are not too close together.

Finally, it's really good at generating fictional characters without using LORA.

It's amazing. I wonder how many characters and artist styles are involved in. and sometimes the saturation is a bit high.

非常棒的模型。我很好奇这里面包含了大约多少角色和画师。请问有角色表或者画师表之类的吗。

有时候感觉色彩过于鲜艳，我还经常能看见微博的水印。

请问要是我训练Lora的话是选这个还是选Illustrious-XL好一点

So good. In my own experience, IL models will usually be very instability in aesthetic when you use artist styles, its style will change a lot even a minor changes to your prompts. And they are usually very likely to have uncontrolled watermarks and texts on pure colored background.

But this one is one of the most stable IL models I used, and the most stable version might be V6 (like composition of a picture or gestures of charactors will be more stable with changing seeds). V5 have more variety, sometime can be better to restore the artist style than V6, at the expense of little bit more instability comparing to V6.

Some artist tags will be weaker in V6. Personally, if your preferred artist tags are active and strong in V6, it will be better. Otherwise V5 will be better.

By the way, V7 seems unstable comparing to previous version.

V5 looked better (in the most tests I tried)

nice model!