Sign In
homemodelsimagesvideospostsarticleschallengeseventsupdatesshop

Oh My Goddess!/ああっ女神さまっ - Manga Art Style (90s-2000s) - Kōsuke Fujishima/藤島 康介

572

Updated: Mar 26, 2026

stylemangaoh my goddess!belldandyskuldah! my goddess

Download

1 variant available

SafeTensor

109.15 MB

Verified:

Type

LoRA

Stats

448

572

235

Reviews

(66)

Published

Mar 22, 2026

Base Model

Illustrious

Hash

AutoV2
B544C3C0E1

Trigger Words

amgstyle
blldnd, ponytail, sidelocks, facial mark, jewelry, earrings, gold collar, fingerless gloves, multiple rings, blue dress, layered dress, off-shoulder dress, stirrup footwear, anklet
rdchar, dark-skinned female, facial mark, very long hair, white hair, gold choker, elbow gloves, single leg pantyhose, jewelry, multiple bracelets, purple dress, plunging neckline, cleavage
prthchar, facial mark, very long hair, low ponytail, jewelry, silver earrings, silver bracelet, multiple bracelets, silver choker, black swimsuit, stomach, sleeveless, cleavage cutout, leather gloves, fingerless gloves, leather belt, vest, leather boots
skuld \(aa megami-sama\), facial mark, black hair, ahoge, very long hair
skuld \(aa megami-sama\), facial mark, brown hair, twintails, ahoge, very long hair
hild \(aa megami-sama\), dark-skinned female, headdress, (very long hair), facial mark, spiked hair, sidelocks, low ponytail, tsurime, short twintails, jewelry, feather earrings, multiple rings
fujimi chihiro, short hair, stud earrings
morisato megumi, brown hair
lndchar, circle facial mark, blue hair, asymmetrical hair
default creator card background decoration
Second Birthday Badge
ilibibu's Avatar

ilibibu

Created on Civitai

After one week, 18 trainings, and 14,719 buzz spent, I have a Lora of my favorite manga art style (hopefully I picked the right training lol). It doesn't capture it perfectly, and hands are a bit unstable sometimes, but I feel it's as good as I'm gonna get for a simple Civitai-trained Lora with a pretty mediocre dataset. I tried to make it look as good as possible without any special settings or prompts, maybe some settings will make it look better.

Trained on 58 stylistically-consistent and hand-tagged images of the main characters from the 90s to the 2000s. Not all images were high quality, but it's hard to find high quality scans of this stuff.

Generation advice:

I put the "m0dernstyl3" tag on images of the high quality cover scans. Adding it to the prompt makes it look slightly more like them.

For some reason it likes to add straps or sleeves to Belldandy's dress, try messing with the negative prompt to avoid that.

Put "torn clothes, broken collar, broken bracelet" in the negative prompt. One image had Belldandy with broken jewelry and ripped clothes. I probably should've removed it, but it's fine if you put those tags in the negative.

Also put "blue eyes" in the negative when generating any character that isn't Belldandy, Urd, or Lind. It should mostly handle eye colors correctly without any tags in the positive prompt.

All the characters generate pretty well except Lind, she only had 2 images. She doesn't look terrible, but not great either. Also Megumi is hit or miss.

Adding "traditional media, scan" might make it look slightly more authentic.