Sign In

Urd (Aa Megami-Sama / Oh My Goddess)

297
1.7k
3.7k
19
Verified:
SafeTensor
Type
LoRA
Stats
1,231
3,699
Reviews
Published
Apr 22, 2023
Base Model
SD 1.5
Training
Steps: 2,500
Epochs: 5
Trigger Words
urd \(aa megami-sama\)
Hash
AutoV2
2D4BD3F3B7
Dark Comic Lips
Deto15's Avatar
Deto15

V2 Description

This version is in my experience much superior to V1. When I trained V1 I haven't put much effort, and since my dataset was small anyway I was content with meh result. Here I've used same set on training images but I've put much more effort into it.

Important thing to note is that this model outputs can differ in style quite a lot even with same prompt and different seeds. They can range from semi-realistic to old school. Personally I love the old school style and I found out you can control it fairly consistently by using retro artstyle tag. My preview images have all been made with weight 0.9 and with retro artstyle put either in prompt or negative prompt, so you can see what a big different in makes. Whichever you prefer, I advice to put it in one of two prompts for more consistent results.

Also retro artstyle have more accurate dark skin, without it it can be a bit lackluster.

Apart from that, prompt is pretty much the same - urd \(aa megami-sama\), facial mark, forehead mark, elbow gloves, choker, bracelet, cape, purple dress, high heels

Cape often seems to need more weight, but that's about it.

V1 Description

Lora model of Urd from Aa Megami-Sama. Previews were made with weight 0.85.

I've been spending too much time on SD so now I wanted to make something simplier and low effort and this is it. There's one outfit and have to struggle a bit for more unusual poses. Details are not very accurate because training set was small. Still for all it's faults, can make nice images also big booba important.

Prompt for right outfit is urd \(aa megami-sama\), facial mark, forehead mark, elbow gloves, choker, bracelet, cape, purple dress, high heels