Sign In

Junko Enoshima 江ノ島盾子 | Danganronpa ダンガンロンパ - v3.0

541
3.4k
60
Verified:
SafeTensor
Type
LoRA
Stats
3,418
0
Reviews
Published
May 27, 2023
Base Model
SD 1.5
Trigger Words
junko enoshima
Hash
AutoV2
A000772F9B

(PS: Civitai keeps tagging Junko's default costume as Sexy Attire and restricting my images, so I had to generate an image with no cleavage and smaller breats to use as cover, lul)

Character LoRA of Junko Enoshima from the Danganronpa series. This is my first LoRA ever and I am pretty satisfied with the results. Version 1.0 was a test run and 2.0 ended up badly, so 3.0 is the first public one.

This LoRA is and will always be free and the images are not permitted to be used for commercial purposes, let's keep it free and available to everyone. You don't need to credit me in any way, but I would be happy if you comment and share some of your results. More Junko is always appreciated :)

Prompt Engineering

TL;DR: Check the gallery for varied examples of how to prompt the positives and negatives.

I'm not very skilled at prompting., but after 500 or so generations I noticed some useful things:

  • The main tags are <lora:junko-enoshima-lora-3-10:1>, junko enoshima, blonde long hair, twintails, blue eyes, bangs. These are intrinsic to the character and it's a good idea to always use them.

  • If you want her default and iconic clothes, the tags are bear hair ornament, nail polish, red nails, red bow, black shirt, underwear, choker, black bra, cleavage, collarbone, sleeves rolled up, white necktie, pleated plaid red skirt, school uniform. You can also add black cross-laced knee boots if you want her to have her default footwear, but it's not necessary on face/upper body images. You can also experiment with generating images of her in different clothing, which can yield nice results with a little luck and good prompting. There are some examples in the gallery.

  • Most of the training images were of Junko's upper body. To recreate this pose you can use the tags 1girl, large breasts, solo, looking at viewer, upper body.

  • Also, the majority of training images had her with the usual open-mouthed smile. To get this default facial expression use the tags open mouth, smile, blush, upper teeth only.

  • Using shiny, shiny hair, high contrast, vibrant colors can get you some colorful results and visually interesting images. Just keep in mind that "high contrast" might generate some weird color palettes.

  • You have the freedom to generate both SFW and NSFW images, this LoRA is highly effective in both.

  • I primarily used the FoxyNSFW V3 model and Anything V4 vae, but I also achieved great results with various anime-based models (and even some realistic ones).

  • And a general warning, this LoRA has a high bias towards abstract/pop-art backgrounds and the color pink in general. If you want landscapes and such you need to specify them in the positive prompt.

But you can also get very good results with different clothing, body positions, facial expressions with some good prompting and a bit of luck. The gallery has some examples of Junko wearing other clothes, having an angry or embarassed expression, etc. You generally want to put her default clothing tags and expressions in the negative prompt and your desired ones in the positive prompt (and pray for a good seed).

About using this LoRA with other LoRAs, Textual Inversions and such. It's kinda of a mixed bag. I would say that about 40% of all combinations I tried were horrible. I'm not sure if it's the LoRA's fault or my lack of prompting skills. That doesn't mean it's impossible, the gallery also has some examples of good combination results. I guess some patience and tinkering with the prompt and weights are needed.

I'm not planning to make any other LoRA models for now, but I'm free for suggestions.