Download
1 variant available
bf16 SafeTensor
Gate_Demi-Humans_Krea2_v2.safetensors
BF16, good balance • 372.6 MB
Verified: 18 hours ago
400 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
20 1 2 3 4 5 6 7 8 9
130 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
(3)
Aug 30, 2026
Known issues
1) Limited style range. I probably overtrained it. Try lowering
strength to 0.8.
2) Trained on base Krea 2 Raw. It works with Turbo models, but not
with heavily finetuned ones. If you are using a heavily modified
model, raise the LoRA strength to 1.3.
3) Tyuule often comes out with a second set of ears. Ear positions
and the jewellery she wears on them are not tagged in the dataset,
so the model averages the different poses together.
Show more

8690 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9
7590 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 90 1 2 3 4 5 6 7 8 9

Krea 2 version
This LoRA generates the demi-human characters from Gate JSDF in Krea 2
Characters: Delilah, Tyuule, Norra, Persia
The minimal prompt needed to summon a character:
Delilah, bunny_girl
Tyuule, red eyes, bunny_girl
Norra, wolf_girl
Persia, cat_girl
The model also picked up some of the manga's art style. Trigger word:
satoru_style
You can also use the race tags bunny_girl, wolf_girl and cat_girl on
their own, but then you need to describe your character's appearance
yourself. With no description the tag will give you one of the named
characters of that race; with a description it gives you your own
character, group scenes included.
Known issues
1) Limited style range. I probably overtrained it. Try lowering
strength to 0.8.
2) Trained on base Krea 2 Raw. It works with Turbo models, but not
with heavily finetuned ones. If you are using a heavily modified
model, raise the LoRA strength to 1.3.
3) Tyuule often comes out with a second set of ears. Ear positions
and the jewellery she wears on them are not tagged in the dataset,
so the model averages the different poses together.
Illustrious version
This LoRA allows you to create images of the Beastmans from "Gate: Thus the JSDF Fought There!"
The dataset consists mainly of manga images, because I was mostly trying to reproduce a beast-folk style.
Lora knows the characters Delilah, Tyuule, Persia, and Norra.
It also understands the terms:
Rabbit_girl,
cat_girl,
wolf_girl,
beast_girl - makes it generate a beast-person just like in manga.

