honey Mix Girl

This is a free/open available Mixed Lora.

If you like my work, feel free to buy me a coffee at ko-fi. https://ko-fi.com/mixboy

The following LoRAs are used. All credits go to the original authors. Thanks!

LoRAs have been merged with the SuperMerger extension

• No.175 Girl by @dawn66666666

• No.179 Girl by @dawn66666666

• No.160 Girl by @dawn66666666

• Dr.She v1.0 by @stablecolab2023

• KoreanDollLikeness (v2.0) by @Kbrloras

Design points

Honey Mix Girl has a face with a more natural look and less makeup, for a more realistic appearance.

• mix No.175 Girl and No.179 Girl

  • both are unique and sexy

• mix No.160 Girl LoRA background brightness

• add a bit more natural style LoRA like Dr.She - Dr.She is a nice example. it is not so pretty and not so cute but well made.

• add high-quality LoRA a bit like as KoreanDollLikeness. it makes LoRA more robust, it has cleaner skin nice smile, etc.

  • if you want to add more makeup and cleaner skin, add "<lora:koreanDollLikenessV20:0.2>" line at the end of prompt.

• you may need to use OUTALL weights preset only - so many low-quality LoRAs have not optimal IN* weights and it makes trouble. simply use only OUTALL weights to solve this issue.

Merge Recipe

Using LoRA Block Weight extension I've tried to find a good mix ratio as follows:

• No.175 : No.179 : No.160 : Dr.She : koreandolllikenessV20 = 0.3 : 0.4 : 0.15 : 0.1: 0.01

• No.175 , No.179 and No.160 are remade its dimension to 128. (the remade dimension version is named NO175, NO179, NO160 in each case.)

• use only OUTALL weights like as
  <lora:NO175:0.3:OUTALL><lora:NO179:0.4:OUTALL><lora:NO175:0.15:OUTALL><lora:Dr.She:0.1:OUTALL><lora:koreandolllikenessV20:0.01:OUTALL>

• use "merge to checkpoint" method supported by the SuperMerger extension

  • fill in the LoRA form: "Dr.She:0.1:OUTALL,NO160:0.15:OUTALL,NO175:0.3:OUTALL,NO179:0.4:OUTALL,koreandolllikenessV20_v20:0.01:OUTALL"

  • output file name as "MERGE1"

  • select checkpoint-A "v1-5-emaonly"

  • then press "merge to checkpoint" button and you can get the "MERGE1".

  • select checkpoint-B "MERGE1", output file name as "new-lora"

  • then press "Make LoRA (alpha * A - beta * B)" button and you can get "new-lora" (alpha == beta == 1)

  • to make "<lora:honeyMixGirl_V1:0.6>" is the default weight, scale the LoRA by "1.66666666666666666"(1/0.6=1.666...) with the "same strength" option.

• now the honey Mix Girl V1 obtained!

In this case, only "OUTALL" weight blocks were selected to reduce possible interference by Lora "IN*" weights. (OUTALL is defined by Lora Block Weight extension as "OUTALL:1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1" that means BASE,OUT03-OUT11 weights are used.)

How to use

• the recommended ratio is 0.5~0.7 like as

  • <lora:honeyMixGirl_V1:0.6> (0.6 is the designed ratio or below 0.6)

• lower ratio weights could be a good starting point for non Lora-friendly-Models.

• the higher ratio could result in overfitted behavior. in this case, lower your ratio.

What's the difference

and what are the advantages of this unlike anything other LoRA girls?

• it is not a trained LoRA, it is a mixed LoRA.

• its recipe is open, anyone who wants to change the mix ratio and wants to get proper results can change their recipe based on it.

• this work is done with the ChilloutMix, and also works fine with ChillyMix or Muse_v1, or other LoRA friendly models.

Recommends

• You can use ADetailer or DDetailer to enhance face details.

• default Clip skip: 2, ETA noise seed delta: 31337

  • for BRA v5 case, Clip skip:1 is preferred.

Useful AUTOMATIC1111's webui extensions

• SuperMerger extension

• Lora Block Weight extension

• webui model toolkit to examine Loras and models

• civitai extension to add several useful features for civitai

License

Like as ChillyMix, its license has no commercial restriction. it is open and free:

• Dr.She and KoreanDollLikenessV20 could be merged and the merges could "Have different permissions when sharing merges" license.

• No.175 Girl , No.179 Girl , No.160 Girl are free! thanks again!

CreativeML-Open Rail-M: This model permits users to:

✔Use the model without crediting the creator

✔Sell images they generate

✔Run on services that generate images for money

✔Share merges using this model

✔Sell this model or merges using this model

✔Have different permissions when sharing merges

Support me

If you like my work, feel free to buy me a coffee at ko-fi. https://ko-fi.com/mixboy