Toko Fukawa , Genocide Jack ( Danganronpa ) [Flux]

This lora should let you create images with either Toko Fukawa or Genocide Jack. Additionally, it provides a few different art styles. Note that I used the name "Genocider Sho" instead of "Genocide Jack" in this lora. Warrented or not, I had some concerns that the name "Jack" might contaminate the training. So Genocide Jack won't be recognized in the prompts and you should isntead use Genocider Sho to reference her.

I trained using natural language for the captions rather than tags so you'll probably get the best results going that route with your prompts. I trained a few specific image types which you can replicate, to varying levels of success. I tried to include a full range of the prompt options in the images. But the following is a brief summary.

Danganronpa the Animation: Use 'This is an anime screencap from Danganronpa The Animation' in the prompt to emulate the art style from the anime series Danganronpa the animation. I've had some success applying this as a general style to non-toko subject matter by lowering the lora strength to 0.9 or below. I usually use a strength of around 1.4 if I'm actually using Toko or Genocide Jack in an image. The Hatsune Miku images I uploaded should give a good idea of what you can do with this lora as far as using it for style.

Sprites from Danganronpa 1 and Danganronpa S: Mention 'danganronpa sprite style' in the prompt to try emulating the art style of Danganronpa 1 and Danganronpa S. For example, 'this image is in the danganronpa sprite style'. Despite the name, I trained on full screenshots, with and without the GUI, with this description along with character sprite rips. So in theory it should be able to 'danganronpafy' a full image.

Clothing: I trained on images of Toko and Genocide Jack from Danganronpa 1, Danganronpa S, Danganronpa the Animation, and Ultra Despair Girls. However, there wasn't enough material from Ultra Despair Girls to translate into consistent results for the lora. You can try different clothing options with prompts like the following:

Danganronpa 1 and Danganronpa the Animation: "Toko Fukawa is wearing her uniform from Danganronpa 1 and her skirt is ankle length. She is wearing a dark purple sailor-style uniform, a white collared shirt, and a red neck ribbon.", "Genocider Sho wearing her uniform from Danganronpa 1 and her skirt is ankle length. She is wearing a dark purple sailor-style uniform, a white collared shirt, and a red neck ribbon. She is holding two pairs of silver scissors, one in each hand. Her tongue is sticking out of her mouth and she has a sinister smile on her face with glowing red eyes."

Swimsuit from Danganronpa S: You can push the image style between anime and game by specifying either 'anime screencap from Danganronpa: The Animation' or 'danganronpa sprite style' when using her swimsuit design. In theory at least. To use her bathing suit the prompt should be something like this, "Toko Fukawa is wearing her swimsuit from Danganronpa S. The swimsuit consists of a purple bikini top. The top is a halter style that ties behind the neck. The bottom part of the swimsuit is a purple wrap. Toko has long dark purple hair in two braids behind her, grey eyes, and large wire-rimmed round glasses."

Ultra Despair Girls: "Toko is wearing her purple school uniform from Danganronpa Another Episode Ultra Despair Girls and her long purple hair is loose. Toko has several hairclips in her hair." Again, I've found that this lora just doesn't do a very good job emulating things from Ultra Despair Girls. It's too little material in the training data to be reliable, but enough that every now and again you'll randomly see some damage to Toko's clothing even when you don't prompt for it. It was an experiment, and sadly the experiment failed.

You might get the correct clothing for Toko/Genocide Jack by just mentioning 'uniform from danganronpa 1' or 'swimsuit from Danganronpa S'. However, it's generally best to be specific and at least give a vague description of what she's wearing, hair style, etc. It doesn't need to be super detailed, just enough to nudge things in the right direction. With a high enough strength on the lora the influence will push things that would normally be 'kinda' right into the correct choices.

Other characters: I also trained on some images of Toko/Genocide Jack with Byakuya Togami, Komaru Naegi, and Monokuma. But that too was a failed experiment. You might be able to get them in the image with careful prompting and a lot of luck. But I think it's too unpredictable to be worth it.

Full GUI: I included a small selection of direct screenshots from Danganronpa 1 in the training data, with all the GUI elements, to see if Flux would be able to replicate it. I might try it again with more images, but for this experiment I consider it another failure. However, you might want to play around with it. Here's the general prompt format that I trained on for Danganropa 1 style screenshots: "Toko Fukawa is wearing her standard sailor school uniform from Danganronpa 1. She is wearing glasses. Toko has purple hair that is in braids and she is facing the camera. The image is a screenshot from Danganronpa 1 in danganronpa sprite style with danganronpa 1 GUI screenshot style. The scene is set on a beach at sunset with a single palm tree on the far right and the ruin of a sunken pirate ship in the water. The school mode setting is visible at the top. Text says "School mode" to the left of a yellow circle with a white smiley face. "Sunset" is written next to the circle. A yellow GUI box in the top right has a music note, “BGM” in black, and a volume bar that has the volume all the way up. The background includes a green message at the bottom of the image that says "Would you like to give Toko a present?". There are two yellow boxes in the foreground, the top has white text that says "Sure, I suppose." The text in the bottom box is white and says "Nah". This is a screenshot from danganronpa 1 in danganronpa sprite style with danganronpa 1 GUI screenshot style. Textbox: "'Want to give Toko cool stuff she'll hate?'", Choicebox: "'Sure, I suppose', 'Nah'", Modetext: "'SCHOOL MODE'", Timetext: "'SUNSET'", Choicebox colors: "'yellow, grey'", Textbox color: "'white, black'"."

Scissors: Flux doesn't seem to have taken to Genocide Jack's scissors very reliably. She might wind up holding them wrong, they might come out mangled or blurry, etc. Run enough generations and you'll eventually get them looking right. But in my experience it seldom happens on a first try. I just referred to them as scissors in the training data so nothing special is needed in the prompts to bring them up.

Locations: I didn't try to train on any specific locations, but some of them appeared enough that you might be able to improve your results by specifying areas in Hope's Peak like "Hope's Peak's library". The physics lab, pool, dorms, Toko's room, art room, martial arts dojo, laundry room, and teacher's lounge all had a number of images from both the anime and game labeled in the training data. Even just mentioning Hope's Peak might be enough to help push the lighting in an image in the right direction if you're generating something set in the school.

Her skirt: It's generally a good idea to specify that she has a long or ankle length skirt. Only a small subset of the training data has her full body in it so the lora doesn't have a very good understanding of what her skirt length is by default. Likewise the victim tally on her thigh is almost always going to be absent.

And that's about it for this one. More Flux Danganronpa sprites to follow.