V4

I added some of his recent works to the V3 dataset (tagged as 'recrbqinori') and made only a few minor tag changes.

There’s not much difference from V3, and I'm not sure which one is better, but I'll be using V4.

V3

I recommend this version. (Based on Illustrious)

I think this is the best version yet.

This LoRA is trained on the noobver dataset (~400 images). Since his early works have different style from latest works, I added three tags as style triggers: pre/mid/post + rbqinori.

The "mid" style appears quite similar to "pre," but "post" has a noticeably different look. While the dataset tags in V3 are only slightly different from noobver, the results are significantly improved.

This version accurately distinguishes smooth, flesh, bloody, and roasted stumps, with better depiction of roasted cases. It also improves inset and exposed brain tags, along with many other specific tags.

However, some limitations remain:

double/triple/quadruple amputee tags don’t work as well, but alternatives like armless amputee, ~~ arm/leg stump, single leg, one arm can achieve similar results.

short/medium/long amputee tags still don’t work.

Eye/ear ???? and reverse ??????? work well, but head ???? doesn’t. The model attempts it but often results in eye or ear ???? instead. :C

For more detailed tags, check the attached ZIP file.

NoobAI

I made this to use on NTR XIII.

Even though it was trained on noobai, I think images generated with noobai look worse. I've included 2b images for comparison.

I tried to create a large LoRA to cover many concepts, but the result is a bit disappointing. Some tags in the dataset can cause terrible or useless effects.

Additionally, since the dataset includes more of his older works, the style is quite different from the previous ones.

Even so, it does differentiate bloody, flesh, smooth, or metal stumps. You can check the exact tags in most of the attached texts.

I don’t know why, but it makes the ????? look a little strange.

I recommend using it alongside other quality LoRAs or artist tags.

V2

Much better lora. I recommend this one.

I add trigger word: rbqinori

You can use most of the tags used in previous version.

And I added character tag [inori yuzuriha (guilty crown)], So It'll be more flexible, I think.

I think it's more like a concept lora, seriously. It does well on specific concept, just like artist's works you know.(You can see them on his pixiv) It makes almost similar pics(at least I think), when using the tags used on datasets.

It might strongly respond to ????????, rolling eyes, etc.

In the use of concept lora, it's great on NoobAI too.

trained on civitai, illustrious.

+ I added .txt sets zip file. trained tags

PONY

same datasets, same parameters, but for pony user.

I prefer ilxl version. You'll need to use different weights on different models, some of the models I tried make disappointing results. But this lora will make quite good works about specific description(like 024.txt of the dataset from V2 version).

V1

rbqinori style lora.

No trigger word.

Because of the datasets, this lora is more like a concept lora. EVEN without any ???? trigger like [????, amputee, blood], this lora can make some ???? stuff,

If you use this for concept lora for some ???? stuff, you can use

triggers above or [disembodied ___, severed ___], something in underscore(___). And it might be better if you higher CFG scale.

And since most of his works are pics of Inori yuzuriha from guilty crown, you'll get pics of her when you use just [pink hair, long hair] or even none of them. I'm sorry I forgot to tag the character. I recommend you to use other prompts depicts character you wants.

preview images are made using Illustrious-XL_SmoothftSPO