This is a free/open available Mixed Lora.If you like my work, feel free to buy me a coffee at ko-fi. <a target="_blank" rel="ugc" href="https://ko-fi.com/mixboy">https://ko-fi.com/mixboy</a>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/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/26124/koreandolllikeness-v20">KoreanDollLikeness (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/65638/no177-girl">No.177 Girl</a> by <a target="_blank" rel="ugc" href="https://civarchive.com/user/dawn66666666">@dawn66666666</a></li></ul><h2 id="heading-154">Design points</h2><ul><li>don't try to make it too beautiful/cute/sexy<ul><li>No.177 Girl is too cute but has a nice style.</li></ul></li><li>preparing a natural style LoRA like Dr.She - 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 as KoreanDollLikeness. it makes LoRA more robust, it has cleaner skin nice smile, etc.</li><li>you may need to use OUTALL weights preset only - so many low-quality LoRA have not optimal IN* weights and it makes trouble. simply use only OUTALL weights to solve this issue.</li></ul><h2 id="heading-155">Merge Recipe (V1)</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>the obtained and used ratio is Dr.She : koreandolllikenessV20_v20 : No.177 = 0.2 : 0.1 : 0.6</li><li>use only OUTALL weights like as "&lt;lora:Dr.She:0.2:OUTALL&gt;&lt;lora:koreandolllikenessV20_v20:0.1:OUTALL&gt;&lt;lora:sd-No.177-000014:0.6:OUTALL&gt;"</li><li>in this case, No.177 Girl is 32 dim, but SuperMerger can "remake dimension" to 128. (the remaked dimension version is named NO177 in this case.)</li><li>merge with the <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger">SuperMerger extension </a>"Dr.She:0.4:OUTALL,koreandolllikenessV20_v20:0.2:OUTALL,NO177:1.2:OUTALL" (x2 multiplied and check "same strength" option.)<ul><li>it's not easy to reproduce the procedure but in this case "same strength" is used.</li></ul></li><li>then merge and obtain the Sweety Mix Girl V1</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="heading-155">Merge Recipe (V2)</h2>Use merge to checkpoint menu instead of Merge LoRA method.<ul><li>same as V1: Dr.She : koreandolllikenessV20_v20 : No.177 = 0.2 : 0.1 : 0.6</li><li>same as V1: use only OUTALL weights like as "&lt;lora:Dr.She:0.2:OUTALL&gt;&lt;lora:koreandolllikenessV20_v20:0.1:OUTALL&gt;&lt;lora:sd-No.177-000014:0.6:OUTALL&gt;"</li><li>same as V1: in this case, No.177 Girl is 32 dim, but SuperMerger can "remake dimension" to 128. (the remaked dimension version is named NO177 in this case.)</li><li>V2: use merge to checkpoint menu with 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: "Dr.She:0.2:OUTALL,koreandolllikenessV20_v20:0.1:OUTALL,NO177:0.6: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:sweetyMixGirl_V2:0.6&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 V2 obtained!</li></ul>※ "merge to checkpoint + "Make LoRA (alpha A - beta B)" method is a much more reproducible method than the"make LoRA" menu.<h2 id="heading-156">How to use</h2><ul><li>recommended ratio is 0.5~0.7 like as<ul><li>&lt;lora:SweetyMixGirl_V1:0.5&gt; or &lt;lora:SweetyMixGirl_V1:0.6&gt;</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="heading-157">What's the difference</h2>What are the advantages of this unlike anything other loRA girls?<ul><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>works nicely with the ChilloutMix, also works fine with ChillyMix or Muse_v1</li></ul><h2 id="heading-158">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 is 2, ETA noise seed delta: 31337<ul><li>for <a target="_blank" rel="ugc" href="https://civarchive.com/models/25494?modelVersionId=63786">BRA v5</a> model, Clip skip:1 is preferred.</li></ul></li></ul><h2 id="heading-159">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="heading-160">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/65638/no177-girl">No.177 Girl</a> is free!</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="heading-3">Support me</h2>If you like my work, feel free to buy me a coffee at ko-fi. <a target="_blank" rel="ugc" href="https://ko-fi.com/mixboy">https://ko-fi.com/mixboy</a>

sweety Mix Girl

2023-06-07 - updated version released

Files with SHA256: d891c1195b670cd76ebdcf7ce5c8f196f453def0ca3628289b8fc16781afd4dd