Great model! I look forward how it will goes. Keep up good work!

AMAZING,it's working better than noobai for my using

Very well.

But is here any dataset difference between versions?

Can you update to v7 for the civitai generator as well?

大佬牛皮

IL v7 is excellent. It's the first IL version I've seen where I don't feel like I'm missing anything I needed from Pony.

v7 is amazing 🔥

In a sense, this series is the best series of Il checkpoint, because many people use it to mix other checkpoint.

V7 is more NSFW and fun than my favorite V5, it reflects LORA and artist tags well, and the backgrounds are beautiful.

I am experimenting with Hires.fixes as eye and body details tend to fall apart.

I'm curious as to why the prompts for the Authors example are arranged the way they are. They are in fact affecting the end result, but I wanted to know why. for example one of the example is as such:

1girl, solo, beige tactical plate carrier, expressionless, blonde hair, yellow hair, white thigh highs, sports sneakers, beige pleated skirt, beige formal blazer, aiming small assault rifle, white gloves, green eyes, thigh gun holster,falling, floating, cityscape, aiming at viewer,

,masterpiece,best quality,amazing quality,

the quality tags are spaced away from the actual main body of the prompt. When I decided to bring it in line with the main body, it made the image a bit worst oddly enough. Additionaly the quality tags for both positive and negatives have no space between each coma, which does also seem to make a difference. Im curious to hear why that is.

Excellent quality and compatibility with illustrious V0.1 LoRAs

ai好像不太会画眼睛总是有些问题

I hope you could train it with v-pred in the future

This is amazing and beautiful

Ver7試してみました！これはもう神modelです！

素晴らしい！

Ver 7 在 NSFW 上很強，但經常給一貓耳女.....

頭髮顏色經常錯

I love this because there is no chunk of noob-ai included. I'm not saying that noob is bad in any way, but it is branching off too much into its own version. the base model Ilxl has its own merits.

Hello!

What kind of LoRA can i use with IL models? XL or Pony based LoRA will work or not?

I'm having some trouble getting images of the same quality that I can make with the create function while running this model locally with the diffusers library in Python. In particular the images I create locally have strange coloration issues and outlines. Does anyone have any recommendations on what might be causing this and how to fix it?

I tried to train my own lora with your checkpoint, but the result looks very different from my dataset (original) image https://imgur.com/a/R2UDpDP Why is that?

At first i really love illustrious_pencil-XL model, but after time by time playing with WAI-NSFW-illustrious-SDXL. I think i've found a replacement, by far this is my main anime model right now.

Thank you & good job for this work.

うーん、、

個人的な意見なんですが、

7.0より6.0、5.0の方が良いイラスト出て来ます。

7.0は被写体が少し遠くて、アニメ風のイラストもちょっとリアル感が出過ぎる感じがしちゃうような気がします。

I think the model is very well done, but I have relatively frequent texts and signatures in the pictures. Despite negative prompts.

My most favorite model has been updated again. V7 is really good in artist tag, and better restore the color of artist style.

Though it still have some issue. It is usually generate the image camera far away from objects. And while having a more restorative color style in artist tag, it is more chance to produce the wrong color than previous version, such as green hair cause green eyes and green nipples.

I have made with this model excellent works, thanks for your hard work!

But the generated were perfect in half body, including head, shoulder, breast and hand. The proportions below the waist are seriously out of balance, especially the calves are not the same length and are too thin compared to the whole body. I tried to change the prompts several times but it didn't work. How can I adjust these? Thank you very much!

I’ve made a lot of great things with this and I appreciate the work you’ve put into making it

Best Illustrious Model by far.

If your torso appears longer when using hi-res in civitai, try setting the CFG scale to 10.

I've been playing with this model for the past few days and it's absolutely fantastic. Anatomical errors are very rare. Really good quality and variety of output. The bodies are a little bit more "generic" but will recognize a lot of artist tags similarly to the base model just as well.

This has been an incredible model so far. Amazing results, great anatomy detail, good artist style implementation. So far no other Illustrious base model I've tried has come close.

Spectacular.

illustrious系模型试了很多，个人感觉肢体表现最好的还是这个和paruparu，不过不知道为啥paru作者删除了模型似乎也不会再更新了，今后只能指着这个模型活了

With the resent departure of illustrious I have tested several base models, and found them to be good to fair.

But this has been the best I have tried, the quality of the images is spectacular, if you are looking for a model to use this is the best you can find.

I'm actually blown away by this one. 🥰

Is a specific VAE needed to make it work at its best? Can I install it together with the other SD 1.5 checkpoints? Do I have to change some settings? Thanks to anyone who can help me: I'm quite inexperienced

♥

this is great, it's what finally convinced me to drop pony.

Is illustrious the new king ? It seems insanely accurate with the prompts! I see almost zero anatomic glitches by just the gallery below. Damn!!

Seems like we have a new king in town guys! Flux and SD is both under threat now :D

A serious question.

For anyone using this model and NoobAI-XL -- why would you use NoobAI? I've tried using both and it seems like this model is just superior in every way. Am I missing something?

Is this model finetuned to use latest up to date dataset from danbooru? I heard original Illustrious xl dataset training is only until april 2024 for the danbooru dataset

What does it mean by adding more nsfw material? Is it mean adding and training more dataset compare to original base model?

WAI will you still try to improve this model? Will we see version 8 someday? Or is this the limit of what can be created today?

Update, review underneath: Actually i did get many pony loras to work with this one, just as they are. I tried to use the comma behind the lora line, as this models description tells you to, but it didnt quite work yesterday. Today it does tho, i must have messed something up.

Ive now tried a few IL models, because i wanted to try something different then pony, i never liked the score tags, and how on specific models they either do nothing, or change everything. And this so far seems like the best IL model. I was really hoping to try NoobAI, like probably most others, but i personally saw the literal article on how to use it, to get pretty subpar results so far, and folded. I will wait until at least a 1.0 of their main version comes out before i even try to learn how to use it right. I think i get what theyre trying to do, but until theyre there, this one seems like the closest that already can do more things then most other IL models.

The problem right now seems to me, that Pony is more limited, because they removed artist tags and whatnot, essentially shooting themselves in the foot and slowly bleeding out. Also i heard its terrible to tweak, so IF someone makes a very good model, its often only good in some specific things, and looses much detail when trying to make things like drawn and realism styles work together. But atm you can still get slightly better results for it, on specific things, mainly because of loras. And that like 90% of them by now are made for pony. So far none of the ones ive tried worked with IL, but theyre very specific and some more general loras could work, idk.

So IL (if trained right, like this one and... not many others), seems to be able to get better, more detailed and consistent results with less mistakes, out of the box. Not perfect, but noticeably better imo. But if you want very specific things only loras can add, like some niece characters, pony rn seems like the almost only option. Especially for furry things, specifically trained furry pony models can get you some characters and species better then this one so far, sometimes even without loras. But this has better details, at least for like 2D-2.5D, and prompt comprehension.

I havent tried that many prompts yet, but to give an example: Trying to make a male argonian, in pony gives me a male argonian, if the model is at all trained on it. In every single IL model, it gives me either a human, most times female, sometimes just with a lizard head. If i really specify in negatives that i dont want any human or female, and put weight on argonian, i can get it to also give me male argonians every time, at least in this model here, not so much in others. However, they are more like "generic" lizard people then specifically argonians. They do have argonian properties tho, so i know that they are in the training data too, i think they just get mixed up more in IL compared to pony. So SOME pony models do know better exactly how an argonian looks like.

However, i like how they look here better. In this model they, and other things, seem to almost never have like any mistakes. Even backgrounds, which btw are also much better then in most pony models ive seen. Pony might know how an argonian looks like more, but this one knows that they dont have 2 tails, 3 fingers, scales in weird places, it knows how to do claws and horns much better, even small details on clothing, jewelry, and ESPECIALLY eyes and teeth. Not every IL model did those better then pony models, but this one does. Not needing adetailer sometimes was not a lie, but if i want to redo them for more detail in higher res, even inpainting seems more consistent then in pony, so that most times gets the job done for me.

I think i can say, because people start making more loras for IL, many convert their pony or XL ones too right now, when there are at least almost as many as for pony, pony probably instantly dies. Loras are the one thing keeping them over other model types right now. And when NoobAI really get it together and releases a model that can do everything that they list they want to do, that will probably eclipse all others, AT LEAST for non-realistic styles, and nsfw. Flux will probably still reign supreme for realism, but it cant do nsfw furry stuff... so they have a 80% smaller target audience then IL lmao.

Hi, just curious. Is this model supports natural language prompting?

Love this model

Model of the year, the best model I ever used.

I'm new to this so sorry if this is a dumb question

What resolution are you guys using for image generations? 1024x1024 (and its variants) seem to be the max but i've read that illustrious can do 1536x1536 (https://arxiv.org/html/2409.19946v1) but when I look at public images on here I see most people sticking to 1024x1024.

Any help much appreciated

love this model great work

This model is incredible; the quality of what it generates is amazing!!!

I’d like to ask a question out of ignorance, and I hope someone can help me. When adding a character or an artist’s style, should it be written in a specific way according to the list of characters it has? Or are the automatic suggestions that appear when you type the name enough? I ask because some characters aren’t on the list, but when I type their name, the automatic suggestion for that character appears.

one of the best i've tried in a long time

Just tried this out for the first time so it may have some flaws but damn its anatomy is insanely good