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><a target="_blank" rel="ugc" href="https://civarchive.com/models/71197/no182-girl">No.182 Girl </a>by <a target="_blank" rel="ugc" href="https://civarchive.com/user/dawn66666666">@dawn66666666</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/60936/no170-girl">No.170 Girl</a> by @dawn66666666</li></ul><ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/22139">Dr.She v1.0</a> by <a target="_blank" rel="ugc" href="https://civarchive.com/user/stablecolab2023">@stablecolab2023</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/48363?modelVersionId=52974">TaiwanDollLikeness (v2.0)</a> 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>by <a target="_blank" rel="ugc" href="https://civarchive.com/user/Kbrloras">@Kbrloras</a></li></ul><h2 id="design-points">Design points</h2>Lovely Mix Girl has a pretty face with a more natural look and less makeup, for a more realistic appearance.<ul><li>mix <a target="_blank" rel="ugc" href="https://civarchive.com/models/71197/no182-girl">No.182 Girl</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/60936/no170-girl">No.170 Girl</a><ul><li>No.182 for cute and a smaller nose, No.170 for face style</li></ul></li><li>add a bit more natural style LoRA like <a target="_blank" rel="ugc" href="https://civarchive.com/models/22139">Dr.She</a> - Dr.She is a nice example. it is not so pretty and not so cute but well made.</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/48363?modelVersionId=52974">TaiwanDollLikeness</a> these make LoRA more robust, it has cleaner skin nice smile, etc.<ul><li>in this LoRA, KoreanDollLikeness + TaiwanDollLikeness weights are higher for a prettier face.</li></ul></li><li>use OUTALL weights preset only - remove all IN* specific attributes.</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:<ul><li>No.182 : No.170 : Dr.She : taiwanDollLikeness : koreandolllikenessV20 = 0.4 : 0.2 : 0.25 : 0.3: 0.3</li><li>No.182, No.170 are remade its dimension to 128. (the remade dimension version is named NO182, NO170 in each case.)</li><li>use only OUTALL weights like as &lt;lora:NO182:0.4:OUTALL&gt;&lt;lora:NO170:0.2:OUTALL&gt;&lt;lora:Dr.She:0.25:OUTALL&gt;&lt;lora:taiwandolllikenessV20_v20:0.3:OUTALL&gt;&lt;lora:koreandolllikenessV20_v20:0.3:OUTALL&gt;</li><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: "NO182:0.4:OUTALL,NO170:0.2:OUTALL,Dr.She:0.25:OUTALL,taiwandolllikenessV20_v20:0.3:OUTALL,koreandolllikenessV20_v20:0.2:OUTALL"</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:lovelyMixGirl_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 lovely Mix Girl V1 obtained!</li></ul>In this case, only "OUTALL" weight blocks were selected to reduce possible interference by Lora "IN*" weights. (OUTALL is defined by <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-lora-block-weight">Lora Block Weight extension </a>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.)<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:lovelyMixGirl_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></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 rel="ugc" href="https://github.com/Bing-su/adetailer/tree/main/adetailer">ADetailer</a> rather than <a 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/22139">Dr.She</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/60936/no170-girl">No.170 Girl</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/71197/no182-girl">No.182 Girl </a>are free! thanks again!</li></ul><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: aa5cd03776cac95905f6c6f3076a8b3d5175e5183dad7f4d4a1a5bf236923c56