Qosic Style

A basic style lora trained on the work of Qosic on Danbooru. Should be especially good with material related to Danganronpa both in expanding a base model's understanding of (some) characters and keeping scenes drifting a bit more to canon. Likewise it has some of the more obscure characters. I've even had it give good results with Ryoko from Danganronpa Zero which is pretty underrepresented in the training data of base illustrious models. The basic style pushes to a more detailed and rich anime style with a strength in broader lines and a more vibrant color palette.

The Illustrious version is best used with tags. The Chroma version should have good support for both tags and natural language prompting but will probably do best with a combination of the two at once.

Chroma Style Prompting

Qosic moves between a few different styles so you'll want to deliberately guide the lora with some style prompting. The following are a good start:

The entire piece is rendered with bold outlines and vibrant colors giving it a lively and dynamic feel. The digital art techniques used enhance the overall appeal making it visually engaging and charming.

The artist employs a digital medium to bring out the vivid colors and intricate details of the characters and their surroundings.

The image is executed in a high-resolution photoshop style.

The artwork is rendered in high resolution showcasing detailed shading and highlights that emphasize the textures of the character's clothing and skin. The artistry is clean and precise with a focus on the character’s INSERT DETAIL REQUIRING ATTENTION features and the intricate details of INSERT DETAIL REQUIRING ATTENTION.

The artwork exhibits a high level of detail and a soft gradient color palette created with digital painting techniques to achieve the smooth gradient-heavy background and the intricate detailed depiction of the character.

Danganronpa Characters

To generate Dangaronpa characters with the Chroma lora I'd suggest two main points. First, you can broaden and strengthen the selection by using my Dangaronpa megapack or Rui Komatsuzaki Styl lora alongside this one (both linked to in this lora's resource section). With the strength of the additional lora lowered below 0.5 to help preserve this lora's style while still giving it some more reference material to work with on the character's distinct traits. This lora has a fairly wide selection to work with but it's still only a small fraction of the total. Whether you're using the Chroma or Illustrious version you'll want to include a danbooru style name for the characters and the tag danganronpa_(series) . This lora was trained on, in order of strength in the dataset, the following:

akamatsu_kaede

nanami_chiaki

kirigiri_kyoko

monokuma

saihara_shuichi

ikusaba_mukuro

harukawa_maki

yonaga_angie

fukawa_toko

naegi_komaru

amami_rantaro

shirogane_tsumugi

k1-b0

oma_kokichi

enoshima_junko

hinata_hajime

chabashira_tenko

komaeda_nagito

iruma_miu

celestia_ludenberg

naegi_makoto

kamukura_izuru

tsumiki_mikan

asahina_aoi

pekoyama_peko

tojo_kirumi

yumeno_himiko

momota_kaito

gokuhara_gonta

sonia_nevermind

fujisaki_chihiro

owari_akane

togami_byakuya

koizumi_mahiru

maizono_sayaka

ogami_sakura

hagakure_yasuhiro

nidai_nekomaru

saionji_hiyoko

tanaka_gundham

ishimaru_kiyotaka

kuzuryu_fuyuhiko

owada_mondo

hanamura_teruteru

kimura_seiko

kuwata_leon

shinguji_korekiyo

soda_kazuichi

yamada_hifumi

togami_byakuya_(danganronpa_2)

kizakura_kouichi

magadori_kyoshika

munakata_kyousuke

sakakura_juuzou

yukizome_chisa

ando_ruruka

gekkogahara_miaya

izayoi_sounosuke

mitarai_ryota

towa_monaca

I'd consider Mondo the cutoff point for being in the training data and actually well represented in the training data. Mondo and anyone below him are probably going to be pretty rough compared to those above him.

With the Chroma lora I'd suggest getting descriptions of their clothing and appearence from the character section of the Dangaronpa fandom wiki.

Though with all that said I should probably stress that this isn't a character lora. The style was the main point, and hope for it to be able to accurately create Danganronpa characters without any additional loras was a secondary concern. It's a nice bonus when it works but it wasn't the main thing I was going for. So saying your mileage may vary is an understatement. It should usually work out quite well with Illustrious since the Dangaronpa characters are well represented in most Illustrious models anyway. While results in Chroma are going to be more erratic since Dangaronpa was almost certainly in the training data but not enough to reliably come through on its own. You'll probably notice that in the gallery's example images. It's quite common even with this lora for the Chroma images to 'almost' get there with adherence to the character's look...but not quite. The bear hairclips for Junko not coming out quite right, Mukuro's look getting mixed with Junko's, Chiaki's hairclip being erratic, etc etc.

Versioning

The jump from 1.0 to 2.0 from Illustrious to Chroma loras shouldn't have too much impact on what the lora can do. It mainly reflects a large reworking of the captions to improve potential training results for models with narrative prompts. I'm assuming the Illustrious model wouldn't benefit too much from those changes so I'm waiting for something larger-scale before doing 2.x version for Illustrious.