Upgrade (2023/12/18) : v3.1 is trained with some body dataset, no need to use lora block weight with this version to get the body look (still need to prompt to get one)

Beware! : It's a face lora trained by extremely close-up face cropped dataset, can generate with lora block weight extension to get more stable body look. Will try to train with body in the future. Sorry for your inconvenience.

Hello it's OKGC, kpop girls/actress lora maker. I speak Mandarin, English, and a little Korean. Follow the info below to get better result. Feel free to ask for help.

Recommended checkpoints:

• supashy mix - https://civarchive.com/models/156202/supashymix

  • final version works best (what I'm using), get it on supashy's gumroad to support him please - https://supashy.gumroad.com/l/supashy_mix_final?layout=profile

• Moonfilm Reality 3.0 - https://civarchive.com/models/43977?modelVersionId=73071

• majicMIX realistic v7 - https://civarchive.com/models/43331?modelVersionId=176425

Recommended parameters: (works best in my opinion)

• Trigger word: hwangyeji

• Sampling Method: Euler

• Steps: 50

• Hires. fix Upscaler: 4x_NMKD-Superscale-SP_178000_G - https://huggingface.co/gemasai/4x_NMKD-Superscale-SP_178000_G/tree/main

  • Put it in folder models/ESRGAN

• Hires upscale: 2

• Hires steps: 15

• Denoising Strength (Hires): 0.25

• Clip skip: 2

• CFG scale: 7

• Size: 512x768

• vae: vae-ft-mse-840000-ema-pruned - https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt

  • Put it in folder models/Vae

• Eta noise seed delta (ENSD: 31337)

  • Find it in SD webui/Settings

    

    

If the face comes out badly, use adetailer to fix it.

https://github.com/Bing-su/adetailer

Recommended parameters: (works best in my opinion)

1. 1st Adetailer model: face_yolov8n.pt

   Inpaint denoising strength: 0.25

   

   

2. 2nd Adetailer model: mediapipe_face_mesh_eyes_only

   Inpaint denoising strength: 0.25

   

   

3. If the face detection of adetailer detects the face of other people in the background,

   set Detection model confidence thereshold: 0.8~0.85

   

Used embeddings:

(how to use?: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Textual-Inversion)

• ng_deepnegative_v1_75t - https://civarchive.com/models/4629/deep-negative-v1x

• badhandv4 - https://civarchive.com/models/16993?modelVersionId=20068 (add disfigured in negative prompt may also helps getting better hand)

• bad_pictures - https://civarchive.com/models/23178?modelVersionId=66043

Highly recommended tools: Lora Block Weight - https://github.com/hako-mikan/sd-webui-lora-block-weight

• How to use?

  <lora:lora name:1:lbw=IN02>  

  <lora:lora name:1:lbw=1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0>

• Recommended parameters (name whatever trigger word you want, it's all about the block weight)

  • Body:

    WINGS:1,0,0,0,0,1,0,1,1,1,0,0,0,0,0,0,0

    BODY:1,1,1,1,1,1,1,1,0,0,0,1,1,1,1,1,1

    BODY0.5:1,1,1,1,1,1,0.2,1,0.2,0,0,0.8,1,1,1,1,1

  • Face: (face shape, hair style, eye shape, eye colors, etc.)

    EYE:1,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0

    MOUTH:1,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0

    FACE:1,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0 (which I use for my lora)

    FACE0.5:1,0,0,0,0,0,0,0,0.8,1,1,0.2,0,0,0,0,0

    FACE0.2:1,0,0,0,0,0,0,0,0.2,0.6,0.8,0.2,0,0,0,0,0

  • Hands:

    HAND:1,0,1,1,0.2,0,0,0,0,0,0,0,0,0,0,0,0

  • outfit: (with tags)

    CLOTHING:1,1,1,1,1,0,0.2,0,0.8,1,1,0.2,0,0,0,0,0

  • pose: (with tags)

    POSE:1,0,0,0,0,0,0.2,1,1,1,0,0,0,0,0,0,0

  • palette/color style: (with tags)

    PALETTE:1,0,0,0,0,0,0,0,0,0,0,0.8,1,1,1,1,1

  • Character:

    KEEPCHAR:1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0

  • Background:

    KEEPBG:1,1,1,1,1,1,0.2,1,0.2,0,0,0.8,1,1,1,0,0

  • Reduce overfitting (same as OUTALL）

    REDUCEFIT:1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1