This is a PonyXL all-in-one character LoRA for Danganronpa all series (1+2+AE+V3+anime). It has 74 characters in dataset. I mainly tested it on AutismmixConfetti, it may or may not work on other base models.
Based on some experience from my previous AIO LoRA for COTE, where character name triggers barely learned nothing but character trait combinations (hair color, eye color, hair style, etc.) do the trigger's job, this time I tried to put more emphasis on character name triggers.
The good news is: it worked as design! For many characters you can now almost retrieve them using prompts as simple as character name only.
The bad news is: it carries more pollution and overfit and makes it harder to apply alternative clothes, for example, iruma miu's clothes always look pink and harukawa maki's twintails always look floating.
For now I don't have plan to go through the LoRA design and training again since AIO LoRAs take super long time to train and much effort to prepare. But hopefully for my next AIO LoRA, I can find better balance between them.
There may be few characters not working as good as the others, for example, great gozu and bandai daisaku due to lack of training data and unique character design, and surprisingly, kuzuryu fuyuhiko and k1-b0, I'm suspecting low coherence in their datasets but I may need more investigation on that part.
Character triggers (TBD):
naegi makoto 苗木誠
asahina aoi 朝日奈葵
togami byakuya 十神白夜
ludenberg celestia セレスティア・ルーデンベルク
fujisaki chihiro 不二咲千尋
yamada hifumi 山田一二三
enoshima junko 江ノ島盾子
ishimaru kiyotaka 石丸清多夏
kirigiri kyoko 霧切響子
kuwata leon 桑田怜恩
owada mondo 大和田紋土
ikusaba mukuro 戦刃むくろ
ogami sakura 大神さくら
maizono sayaka 舞園さやか
fukawa toko 腐川冬子
hagakure yasuhiro 葉隠康比呂
akamatsu kaede 赤松楓
saihara shuichi 最原終一
harukawa maki 春川魔姫
iruma miu 入間美兎
tojo kirumi 東条斬美
amami rantaro 天海蘭太郎
gokuhara gonta 獄原ゴン太
hoshi ryoma 星竜馬
k1-b0 キーボ
momota kaito 百田解斗
oma kokichi 王馬小吉
shinguji korekiyo 真宮寺是清
yonaga angie 夜長アンジー
yumeno himiko 夢野秘密子
chabashira tenko 茶柱転子
shirogane tsumugi 白銀つむぎ
hinata hajime 日向創
owari akane 終里赤音
ultimate imposter 超高校級の「詐欺師」
nanami chiaki 七海千秋
kuzuryu fuyuhiko 九頭龍冬彦
tanaka gundham 田中眼蛇夢
saionji hiyoko 西園寺日寄子
mioda ibuki 澪田唯吹
soda kazuichi 左右田和一
koizumi mahiru 小泉真昼
tsumiki mikan 罪木蜜柑
komaeda nagito 狛枝凪斗
nidai nekomaru 弐大猫丸
pekoyama peko 辺古山ペコ
nevermind sonia ソニア・ネヴァーマインド
hanamura teruteru 花村輝々
naegi komaru 苗木こまる
kemuri jataro 煙蛇太郎
utsugi kotoko 空木言子
daimon masaru 大門大
towa monaca 塔和最中
shingetsu nagisa 新月渚
towa haiji 塔和灰慈
hagakure hiroko 葉隠浩子
kamukura izuru カムクライズル
fujisaki taichi 不二咲太市
asahina yuta 朝日奈悠太
monokuma モノクマ
monomi モノミ
yukizome chisa 雪染ちさ
bandai daisaku 万代大作
great gozu グレートゴズ
sakakura juzo 逆蔵十三
tengan kazuo 天願和夫
kizakura koichi 黄桜公一
munakata kyosuke 宗方京助
gekkogahara miaya 月光ヶ原美彩
ando ruruka 安藤流流歌
mitarai ryota 御手洗亮太
kimura seiko 忌村静子
kirigiri jin 霧切仁
izayoi sonosuke 十六夜惣之助
Description
FAQ
Comments (20)
74 in one?
zaaaaamn
Unfortunately I can't get this Lora to work, despite the previous Danganronpa Lora working extremely well for me.
Edit: Got it to work by setting the character info to 1.5, like - (akamatsu kaede, ahoge, purple eyes, blonde hair:1.5)
Thanks for your feedback! May I ask what base model are you using? It would be very helful if you can post some use cases.
Also, you can try 1) moving character triggers and traits to the beginning of your prompts 2) enlarge the LoRA weight 3) different or shortened quality words / negatives
@Akiseki I was using oneFORALLPonyFantasy_v20DPO, and I had already tried increasing the weights, I will try using other checkpoints and move the character traits and get back to you
@Akiseki Seemingly no luck with several autism or pony checkpoints at varying weights.
@carmerkraig843 Can you reproduce my examples? It sounds like the LoRA is not loaded properly. I have tried oneFORALL and CivitAI site generator and the results should not be too far away. I have added them as posts below.
You can try varying LoRA weights with fixed seed to see if the generated results stay unchanged, which means the LoRA is not loaded.
If you are sure you are doing the same thing as other LoRAs, can you see any error message in the console?
@Akiseki If anything the background gets slightly more white/grey, the character pose changes a little bit, but in no recognizable fashion, as I up the weight the image becomes more distorted. I tried adding weight specifically to the name used when calling the character, but that didn't work. And there is no error message.
@carmerkraig843 You can try to reproduce my examples I posted below (oneFORALL example:https://civitai.com/images/20243884). In case you are not using webui, you should get something different but not too far away, if the LoRA is fully activated.
@Akiseki Thanks for the help, I got it working by surrounding the character name and attributes like (akamatsu kaede, ahoge, purple eyes, blonde hair:1.5) and it started to work
@carmerkraig843 I don't think this is normal but I can't reproduce the issue with my webui environment😢
@Akiseki Well with the 'fix' I used the output is as good as your previous danganronpa model and as good as your example pictures, so thank you for making such a great model
nice
Do you do Requests?
You can write down the request in comment, but I can't guarantee whether I will fulfill it. It depends on the workload and how much I like the idea.
I'd Like a Lora like this. but with Ensemble Stars Characters(all of them are boys. except for a girl or something). Sorry if it sounds tiring. but you can note it and do it later. I won't force you to do it now
@Boredafk https://civitai.com/models/630311
I'll try it out later, thank you. Hope you can keep updating it (if you can)
Is there any way you could do a kiyondo one? Just did a Taka and would love to make a Kiyondo one
sorry I have checked that the white hair version kiyondo was not included in the dataset
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.