You can deploy this model and start earning money today!

samjwu/
AinzOoalGown

Name: AinzOoalGown
Availability: InStock

gpl-3.0

Views

No views yet

Ainz Ooal Gown LoRA

Description

Low Rank Adaptation for Ainz Ooal Gown from the Overlord series.

A reference image from the light novel. See other references for the anime/manga in the references folder

Usage

If using Stable Diffusion WebUI:

Download AinzOoalGownLoRA.safetensors
Copy and paste into stable-diffusion-webui/models/Lora
Use the trigger word <lora:AinzOoalGown:0.8> in the prompt. Works best with weights ranging from 0.5 to 0.8 (default is 1)

Examples

The Good

Loses some likeness of the original character, but retains that cool skeletal undead look. See more (good) examples in the examples folder

Parameters:

Prompt: 1boy, (overlord, lich, skeleton, skull, bone, ainz ooal gown:1.2), (best quality:1.2), (masterpiece:1.2), (highres:1.2), extremely detailed, illustration, solo, (looking at viewer:1.2), (sharp focus:1.5), (masterpiece illustration:1.5), light particles, glowing light, lens flare, <lora:AinzOoalGown:0.8>

Negative prompt: EasyNegative, worst quality, low quality, normal quality, blurry, deformed, disfigured, morbid, mutated, bad anatomy, bad art, lowres, polar lowres, bad face, bad hands, bad body, bad feet, bad proportions, bad leg, more legs, gross proportions, poorly drawn, (text:1.5), error, missing fingers, missing arms, missing legs, short legs, extra digit, 2girls, 1girl, long skirt, indoors, out of frame, only upper body, only lower body, petite, print, chibi, only head

Steps: 30, Sampler: DDIM, CFG scale: 7, Seed: 4149917258, Size: 512x512, Model hash: 89d59c3dde, Model: nai

The Bad

Still interesting, but misshapen

The Ugly

It kinda looks like him if you squint

Training Details

Training Model: Anything V3.1
Training Dataset: 13 images
Training Repeats: 10
LoRA Network Linear Dimension: 32
LoRA Network Linear Alpha: 16
Optimizer: AdamW8bit
UNet Learning Rate: 1e-4
Text Encoder Learning Rate: 5e-5
Learning rate warmup steps: 0
Learning rate Scheduler: constant
Sampler: ddim
Epochs: 10
Training Batch Size: 6
Precision: fp16 Mixed
Clip Skip: 2
Gradient Accumulation Steps: 1
Total Steps: 230

samjwu/AinzOoalGown