<h1 id="illustrious-citlali-(genshin-impact)-mmbawfh83">[Illustrious] Citlali (Genshin Impact)</h1><h2 id="disclaimer-qeyfqvv41">Disclaimer</h2><ul><li>Do not use my LoRA to produce AI image and using in official contest.</li></ul><h2 id="description-vdseebgyl">Description</h2><ul><li>UNet + CLIP-L LoRA+</li><li>all images is generated with WAI-IL (832x1216 + Hi-Res Fix 2x)</li></ul><h2 id="trigger-word-55k6w1a0l">trigger word</h2>debut costume (remove some component for upper body, cowboy shot or others camera angle<pre><code>citlali \(genshin impact\), 1girl, solo, pink hair, long hair, blue eyes, pink pupils, facial mark, breasts, braid, forehead mark, pink pupils, very long hair, hair ornament, floating hair ornament, twin braids, jewelry, bracelet, black leotard, feather hair ornament, hair between eyes, black sleeves, uneven gloves, black footwear</code></pre>cape cr. <a target="_blank" rel="ugc" href="https://civarchive.com/user/fansay">fansay</a><pre><code>waist cape</code></pre>any costume <pre><code>citlali \(genshin impact\), 1girl, solo, pink hair, long hair, blue eyes, pink pupils, facial mark, breasts, braid, forehead mark, pink pupils, very long hair, hair ornament, floating hair ornament, twin braids,</code></pre>cosplay recommended (maybe work in some character) cr. <a target="_blank" rel="ugc" href="https://civarchive.com/user/fansay">fansay</a><pre><code>citlali \(genshin impact\) \(cosplay\), jewelry, bracelet, black leotard, feather hair ornament, black sleeves, uneven gloves, black footwear, waist cape,facial mark, forehead mark,hair ornament, floating hair ornament,</code></pre>for empty eyes expression replace blue eyes, pink pupils with empty eyes<pre><code>empty eyes</code></pre><h3 id="limitations-vwd00fdig">Limitations</h3><ul><li>not found (at this moment)</li></ul><h2 id="training-details-f43mv2hqx">Training Details</h2>LoRA rank<ul><li>LoRA+</li></ul>dataset<ul><li>819 images</li></ul>parameters<ul><li>resolution = 1024</li><li>batch size = 4</li><li>network dim,alpha = 16,16</li><li>mix/save precision = bf16/bf16</li><li>optmizer = Lion</li><li>UNet LR = 1.6e-5</li><li>TE LR = 8e-06 (50%)</li><li>scheduler = cosine_with_min_lr (min_lr_ratio 0.25)</li><li>l2 loss</li></ul>steps<ul><li>epochs = 8</li><li>total steps = 3276</li><li>repeat = 2</li></ul>tools<ul><li>kohya-ss GUI v24.4.0 Custom Version by me</li><li>torch 2.5.0 cu124</li><li>Runpod RTX 4090</li></ul>avg weight<ul><li>UNet average weights : 0.010891411538151178</li><li>TE1 average weights : 0.008991780884874364</li></ul>*This LoRA is for studying LoRA training with new technique so do not use for damaging the character (also support her too).*If you see my model reposted on other image generation service please report to me.

Files with SHA256: 7e9a67d49f550e75c3d235b1dc44153524f59d1a37a1c287830533b184b1ae04