This is a free/open available Mixed LoRA.The following LoRAs are used. All credits go to the original authors. Thanks!(LoRAs have been merged with the <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger">SuperMerger extension)</a><ul><li>some pretty but not pretty well-made LoRAs (LoRA name / chosen weight ratio and author)<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/25562/jihyo-twice">Twice Jihyo</a> / 0.3, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/Xelor">@xelor</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/16339?modelVersionId=19292">Minju Angle</a> / 0.2, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/alelele">@alelele</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/25903/kim-minju">Kim minju</a> / 0. 2, by @fingertwist</li></ul></li><li>good LoRAs<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/73379/no183-girl">No.183 Girl </a>/ 0.1, by @dawn66666666</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/63198?modelVersionId=67735">No.175 Girl</a> / 0.1, by @dawn66666666</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/65638?modelVersionId=70288">No.177 Girl</a> / 0.3, by @dawn66666666</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/28811?modelVersionId=34562">JapaneseDollLikeness (v1/5)</a> / 0.3, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/Kbrloras">@Kbrloras</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/26124/koreandolllikeness-v20">KoreanDollLikeness (v2.0)</a> / 0.15, by @Kbrloras</li></ul></li><li>Anime LoRA<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/51791/ririka-momobami">Ririka Momobami</a> / 0.1, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/ForkY">@ForkY</a></li></ul></li><li>Randomly selected LoRAs for style<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/14749/liliya">Liliya</a> / 0.1, by @Arashiooo</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/80327/chinese-doll-idol-round-face">Chinese doll idol Round face</a> / 0.25, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/qiuqiu19147">@qiuqiu19147</a></li></ul></li></ul><h2 id="design-points">Design points</h2>Pretty Mix Girl has a pretty face with a more natural look and feels for a more realistic appearance, by mixing multiple LoRAs<ul><li>mix <a target="_blank" rel="ugc" href="https://civarchive.com/models/25562/jihyo-twice">Twice Jihyo</a> (0.3) + <a target="_blank" rel="ugc" href="https://civarchive.com/models/16339?modelVersionId=19292">Minju Angle</a> (0.2)+ <a target="_blank" rel="ugc" href="https://civarchive.com/models/25903/kim-minju">Kim minju</a> (0.2) for a pretty face.<ul><li>but, mix No.183 (0.1) + No.175 (0.1) + No.177 (0.3) to wash out the original direct appearance personality.</li></ul></li><li>add a bit of anime LoRA <a target="_blank" rel="ugc" href="https://civarchive.com/models/51791/ririka-momobami">Ririka Momobami</a> to add a pretty attribute.<ul><li>a small amount of the ratio of anime LoRA makes LoRA prettier.</li></ul></li><li>add high-quality LoRA a bit, like <a target="_blank" rel="ugc" href="https://civarchive.com/models/26124/koreandolllikeness-v20">KoreanDollLikeness</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/28811?modelVersionId=34562">JapaneseDollLikeness </a>these make LoRA more robust, it has cleaner skin nice smile, etc.<ul><li>in this case, JapaneseDollLikeness weights are higher for eyes,</li></ul></li><li>Randomly selected LoRAs to add some properties<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/14749/liliya">Liliya</a> (0.1) for face shape, <a target="_blank" rel="ugc" href="https://civarchive.com/models/80327/chinese-doll-idol-round-face">Chinese doll idol Round face</a> (0.25) for round face shape, to get an ideal round shape face.</li></ul></li><li>use OUTALL1 weights preset only - remove all IN* specific attributes.<ul><li>OUTALL1:0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1 (BASE == 0) (OUTALL case, BASE == 1)</li></ul></li></ul><h2 id="merge-recipe">Merge Recipe</h2>Using <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight">LoRA Block Weight extension</a> I've tried to find a good mix ratio as follows:&lt;lora:Ririka Momobami:0.1:OUTALL1&gt;&lt;lora:koreandolllikenessV20_v20:0.15:OUTALL1&gt;&lt;lora:JapaneseDollLikeness_v15:0.3:OUTALL1&gt;&lt;lora:xelor_TwiceJihyo_2.5:0.3:OUTALL1&gt;&lt;lora:NO183:0.1:OUTALL1&gt;&lt;lora:NO175:0.1:OUTALL1&gt;&lt;lora:NO177:0.3:OUTALL1&gt;&lt;lora:mjftwv2:0.2:OUTALL1&gt;&lt;lora:kim-minju-23-v1:0.2:OUTALL1&gt;&lt;lora:Liliya_v1.0:0.1:OUTALL1&gt;&lt;lora:Chinese doll idol Round face:0.25:OUTALL1&gt;(in this case, OUTALL1:0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1)<ul><li>use "merge to checkpoint" method supported by the <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger">SuperMerger extension</a><ul><li>fill in the LoRA form: "Ririka "Momobami:0.1:OUTALL1,koreandolllikenessV20_v20:0.15:OUTALL1,JapaneseDollLikeness_v15:0.13:OUTALL1,xelor_TwiceJihyo_2.5:0.3:OUTALL1,NO183:0.1:OUTALL1,NO175:0.1:OUTALL1,NO177:0.3:OUTALL1,mjftwv2:0.2:OUTALL1,kim-minju-23-v1:0.2:OUTALL1,Liliya_v1.0:0.1:OUTALL1,Chinese doll idol Round face:0.25:OUTALL1"</li><li>output file name as "MERGE1"</li><li>select checkpoint-A "v1-5-emaonly"</li><li>then press "merge to checkpoint" button and you can get the "MERGE1".</li><li>select checkpoint-A "MERGE1", and checkpoint-B "v1-5-emaonly", output file name as "new-lora"</li><li>then press "Make LoRA (alpha * A - beta * B)" button and you can get "new-lora" (alpha == beta == 1)</li><li>to make &lt;lora:prettyMixGirl_V1:0.6&gt; is the default weight, scale the LoRA by "1.66666666666666666"(1/0.6=1.666...) with the "same strength" option.</li></ul></li><li>now the pretty Mix Girl V1 is obtained!</li></ul>only "OUTALL1" weight blocks were selected to reduce possible interference by Lora "IN*" weights. (OUTALL1 is user defined setting (used by <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight">Lora Block Weight extension</a> )<a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight"> </a>as "OUTAL1L:0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1" that means set BASE == 0 and only OUT03-OUT11 weights are used.)<h2 id="how-to-use">How to use</h2><ul><li>the recommended ratio is 0.5~0.7 like as<ul><li>&lt;lora:prettyMixGirl_V1:0.6&gt; (0.6 is the designed weight ratio or below 0.6)</li></ul></li><li>lower ratio weights could be a good starting point for non LoRA-friendly-Models.</li><li>the higher ratio could result in overfitted behavior. in this case, lower your ratio.</li><li>when using Dynamic-Thresholding or a high CFG-Scale, lowering the weight ratio will be helpful.</li></ul><h2 id="whats-the-difference">What's the difference</h2>and what are the advantages of this unlike anything other LoRA girls?<ul><li>it is not a trained LoRA, it is a mixed LoRA.</li><li>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.</li><li>this work is done with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/6424/chilloutmix">ChilloutMix</a>, and also works fine with <a target="_blank" rel="ugc" href="https://civarchive.com/models/58772?modelVersionId=63220">ChillyMix</a> or <a target="_blank" rel="ugc" href="https://civarchive.com/models/13564/musev1">Muse_v1,</a> or other LoRA friendly models.</li></ul><h2 id="recommends">Recommends</h2><ul><li>You can use <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a> or <a target="_blank" rel="ugc" href="https://github.com/dustysys/ddetailer">DDetailer</a> to enhance face details.</li><li>default Clip skip: 2, ETA noise seed delta: 31337</li><li>for <a target="_blank" rel="ugc" href="https://civarchive.com/models/25494?modelVersionId=63786">BRA v5</a> case: (WARNING: experimental instructions)<ul><li>Clip skip:1 is preferred.</li><li>try to change your sampling method. for example, DPM2 or DPM2 Karras</li><li>or lower CFG Scale ~4</li><li>or Enable <a target="_blank" rel="ugc" href="https://github.com/mcmonkeyprojects/sd-dynamic-thresholding">Dynamic Thresholding</a> with the following advanced options:<ul><li>Set CFG Scale 7~13 and mimic CFG Scale to ~4</li><li>Top percentile of latents to clamp: ~100</li><li>Mimic Scale Scheduler: Linear Down or what you want</li><li>Minimum Mimic Scale Schedular: ~4</li><li>CFG Scale Scheduler: Linear Down or what you want</li><li>Minimum CFG Scale Scheduler: ~4</li><li>try to change your sampling method. (e.g.) DPM++2M Karras)</li></ul></li><li>use <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer/tree/main/adetailer">ADetailer</a> rather than <a target="_blank" rel="ugc" href="https://github.com/dustysys/ddetailer">DDetailer</a></li></ul></li><li>try to add "&lt;lora:<a target="_blank" rel="ugc" href="https://civarchive.com/models/13941">epi_noiseoffset2</a>:1.8&gt;" LoRA and adjust the weight ratio of 1.8 if necessary</li></ul><h2 id="useful-automatic1111s-webui-extensions">Useful AUTOMATIC1111's webui extensions</h2><ul><li><a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger">SuperMerger extension</a></li><li><a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight">Lora Block Weight extension</a></li><li><a target="_blank" rel="ugc" href="https://github.com/arenasys/stable-diffusion-webui-model-toolkit">webui model toolkit</a> to examine Loras and models</li><li><a target="_blank" rel="ugc" href="https://github.com/civitai/sd_civitai_extension/">civitai extension </a>to add several useful features for civitai</li></ul><h2 id="license">License</h2>Like as <a target="_blank" rel="ugc" href="https://civarchive.com/models/58772/chillymix">ChillyMix</a>, its license has no commercial restriction. it is open and free:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/73379/no183-girl">No.183 Girl</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/63198?modelVersionId=67735">No.175 Girl</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/65638?modelVersionId=70288">No.177 Girl</a> (by @dawn66666666<a target="_blank" rel="ugc" href="https://civarchive.com/models/71197/no182-girl">) </a>are free! thanks again!</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/14749/liliya">Liliya</a> (by @Arashiooo) and <a target="_blank" rel="ugc" href="https://civarchive.com/models/25903/kim-minju">Kim minju</a> (by @fingertwist) are free!.</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/51791/ririka-momobami">Ririka Momobami</a> (by <a target="_blank" rel="ugc" href="https://civarchive.com/user/ForkY">@ForkY</a>) is "no sell model" and it means almost free!</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/28811?modelVersionId=34562">JapaneseDollLikeness </a>and <a target="_blank" rel="ugc" href="https://civarchive.com/models/26124/koreandolllikeness-v20">KoreanDollLikenessV20</a> could be merged and the merges could "Have different permissions when sharing merges" license.</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/80327/chinese-doll-idol-round-face">Chinese doll idol Round face</a> (by <a target="_blank" rel="ugc" href="https://civarchive.com/user/qiuqiu19147">@qiuqiu19147</a>), <a target="_blank" rel="ugc" href="https://civarchive.com/models/25562/jihyo-twice">Twice Jihyo</a> (by <a target="_blank" rel="ugc" href="https://civarchive.com/user/Xelor">@xelor</a>) and <a target="_blank" rel="ugc" href="https://civarchive.com/models/16339?modelVersionId=19292">Minju Angle</a> (by <a target="_blank" rel="ugc" href="https://civarchive.com/user/alelele">@alelele</a>) could be merged and this merge could have different permission when sharing merges too.</li></ul>so It is possible to release it without any restriction!<a target="_blank" rel="ugc" href="https://huggingface.co/spaces/CompVis/stable-diffusion-license">CreativeML-Open Rail-M</a>: 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<h2 id="donation">Donation</h2>Feel free to donate to me if you want to support my work! at ko-fi <a target="_blank" rel="ugc" href="https://ko-fi.com/mixboy">https://ko-fi.com/mixboy</a>

Files with SHA256: eb749197426c9fea03198f86a285389eeacb33dc355a668e3ef45d321db51593