I wouldn't recommend actually using this model as it's mainly for testing out some training ideas to see what sticks. There's some great stable Tsumugi loras on this site, including mine. Still, this one should hopefully offer something a little different, for better or worse.

Models with multiple versions are instances where I'm really just throwing things at the wall to see what sticks. In those cases, I probably couldn't really recommend one of those over another. At the time of writing, that's why there's a bucket of flux versions. I'm not really happy with any of my flux loras in this 'branch', but I don't exactly dislike them either. Of the Flux loras, I'm leaning to Flux 2.01 e15 being the one to try out if you're curious. The e4 has the best gallery images to be sure. But that's more to do with the fact that I used the flux lyh anime model for most of those gens. I think the high gen quality of the lyh model is covering up for some of the issues with my Tsumugi Flux e4 lora. That said, now that I'm doing more testing the e25 is looking pretty good using lyh as well. Sadly flux crawls on my system so testing these is a slow process since it monopolizes my GPU.

Right now the main intent for this "branch" is to try out methods of balancing narrative and tag based captioning to hopefully get the best of both worlds. Along with seeing differences with how various models take to it. This is also part of a larger effort to get my stuff ready to train loras for chroma and get some of my illustrious loras working with flux as well.

I'm starting out by uploading the Illustrious "build" but should be following up shortly with loras for the other models.

That said, if I've uploaded a model here it's 'probably' usable for image generation with Tsumugi Shirogane or Chiaki Nanami from the Danganronpa franchise. Hopefully with both tags and descriptive prompts. How well theory holds out in practice remains to be seen. But in general you should be able to just toss danbooru style tags at it reliably. And narrative descriptions probably won't work as well, but you should hopefully get something. And a combination of the two should be fine as well. Or at least slightly better than you might expect from a base model without this lora. The biggest caveat there is Stable Diffusion 3.5. My results with 3.5 are improving, but it's pretty unpredictable. At least with my workflow.

Training data consists of official, higher quality unofficial , and a smattering of synthetic art of Tsumugi Shirogane with all her official outfits (V3, Danganronpa S, 10th anniversary dress). I've balanced it out with a variety of images from the Danganronpa animes and games to push that style a bit while also hopefully allowing for direction of image generation to go with each style. I've also included a fair amount of data for Chiaki Nanami. Not nearly as much, but enough that it should be able lto reliably generate images of her as well even if a base model doesn't have much or anything in the way of danganronpa knowledge. Likewise it should do a good job of having both of them together in the same image.

The base tags for Tsumugi are:
1girl, solo, danganronpa_(series),danganronpa_v3:_killing_harmony, shirogane_tsumugi,tsumugi shirogane, long_hair, round_eyewear, sidelocks, very_long_hair, wavy_hair, blue_hair,dark_blue_hair, blue_eyes, breasts,medium_breasts

And for her standard Danganronpa V3 uniform:
bow, bowtie, buttons, collared_jacket, collared_shirt, corset, black_corset, dress_shirt, glasses, jacket, black_jacket, lapels, layered_sleeves, long_hair, long_sleeves, notched_lapels, open_clothes, open_jacket, orange_bow, orange_bowtie, pleated_skirt, black_skirt, round_eyewear, shirt, sidelocks, skirt, very_long_hair, white_shirt, shoes, brown_footwear, black_socks

For changing outfits you should be able to just swap out the uniform tags with something else.

To get both Tsumugi and Chiaki with tags, try a prompt like this:

(((2girls))), multiple_girls,side-by-side, danganronpa_(series), BREAK, danganronpa_v3:_killing_harmony, shirogane_tsumugi,tsumugi shirogane, long_hair, round_eyewear, sidelocks, very_long_hair, wavy_hair, blue_hair, blue_eyes, medium_breasts,BREAK,danganronpa_2:_goodbye_despair, nanami_chiaki, chiaki nanami, breasts, pink_hair,medium_hair, flipped_hair, pink_eyes, spaceship_hair_ornament, hair_ornament,large_breasts , danganronpa_(series), danganronpa_v3:_killing_harmony, shirogane_tsumugi,black_socks, blazer, blue_corset, blue_hair, blue_jacket, blue_skirt, bow, bowtie, breasts, brown_footwear, buttons, collared_shirt, corset, glasses, jacket, kneehighs, layered_sleeves, long_hair, long_sleeves, medium_breasts, medium_skirt, open_clothes, open_jacket, orange_bow, orange_bowtie, pleated_skirt, pocket, round_eyewear, shirt, shoes, sidelocks, skirt,socks, very_long_hair, wavy_hair, white_shirt, BREAK, danganronpa_2:_goodbye_despair, galaga, nanami_chiaki, animal_bag, black_jacket, bow, brown_hair, brown_skirt, cat_bag, hood, hood_down, hooded_jacket, jacket, long_sleeves, medium_hair, miniskirt, neck_ribbon, pink_bag, pink_bow, pink_eyes, pink_ribbon, pleated_skirt, ribbon, shirt, shoes, skirt, white_shirt, zettai_ryouiki,BREAK, PUT THE REST OF YOUR PROMPT HERE

Or in their Danganronpa S bathing suits: And here's the prompt for them both in their bathing suits: 2girls,BREAK,Danganronpa S,bow,breasts,chiaki nanami,nanami_chiaki, pink_eyes,pink_hair,cleavage,collarbone,danganronpa_(series),danganronpa_2:_goodbye_despair,danganronpa_s:_ultimate_summer_camp,flipped_hair,frills,hair_ornament,spaceship_hair_ornament,large_breasts,mole, mole_on_breast, white_bikini,white_bikini_chiaki,white_lace, medium_hair,navel,BREAK, shirogane_tsumugi, tsumugi shirogane, danganronpa_v3:_killing_harmony, blue_eyes,blue_hair, orange_bikini,orange_bikini_tsumugi,light_orange_bikini,long_hair,medium_breasts, round_eyewear,sidelocks,stomach,very_long_hair,wavy_hair

My default descriptive prompt is:
Tsumugi Shirogane is wearing her uniform from Danganronpa V3. Tsumugi Shirogane has knee-length dark teal-blue hair parted in the center, which is straight and smooth at the base and gradually splits into waves near the tips. Her eyes are large, round, and are a soft teal-green color. Tsumugi wears round, rimless glasses with white temples. Tsumugi's Danganronpa V3 school uniform consists of a white blouse with an orange bow tied at her collar underneath a two-button, pocketed, teal-black school blazer adorned with two silver buttons on each lapel and a single button on each gorge. She wears a knee-length teal-black pleated skirt with two dark teal stripes lining the hem. Her legwear consists of dark navy knee-length socks and teal-gray loafers.

However, so far I seem to be getting good results with this more concise version:

Tsumugi Shirogane and Chiaki Nanami. Tsumugi Shirogane wears her Danganronpa V3 school uniform - white blouse with orange bow, teal-black blazer with silver buttons, knee-length pleated dark navy blue skirt, navy knee-high socks, and teal-gray loafers. Tsumugi's long knee-length dark teal-blue hair is parted in the center, straight at the base and wavy at the tips. Tsumugi wears round rimless glasses. Chiaki Nanami wears her Danganronpa 2 uniform - white button-up shirt with dark teal-gray placket, light pink ribbon bow, pale beige pleated skirt, dark teal-gray hooded cardigan with gray lines, black thigh-high socks, and light pink Mary-Jane shoes. Chiaki's neck length dusty light pink hair curls outwards with inward-curling face locks, straight bangs split right, and her hair held in front with a white Galaga spaceship hairclip.

Experimental Stuff

You may or may not get extra impact from using some of these prompts with this lora. But like I said, experimental. The positional and "one person cosplaying but not the other" stuff probably isn't represented well enough in the dataset to really have much impact but I was curious. If nothing else the positional tags do seem to help the models understand that there should be "two" people in the image rather than three or more.

komatsuzaki_rui_(style)

official_art, komatsuzaki_rui_(style)

danganronpa sprite style, danganronpa_sprite_style

This is an anime screencap from Danganronpa 3. , anime_screencap_danganronpa3, danganronpa_3_(anime), anime_screencap

This anime screencap is from Danganronpa the Animation. , anime_screencap_danganronpa_the_animation, anime_screencap

glossy surfaces,high contrast, intricate details,masterful shading,shiny skin,smooth shading, modern anime, anime

1cosplayer

2cosplayers

chiaki_left, tsumugi_right

tsumugi_left, chiaki_right

modern, modern anime, 2023, 2024, 2025