<h1 id="illustrious-raora-panthera">[Illustrious] Raora Panthera</h1><h2 id="disclaimer">Disclaimer</h2><ul><li>Do not use my LoRA to produce AI image and tagging her fanart hashtag in Twitter/X or if you have an enough contribution in that image then fine.</li></ul><h2 id="description">Description</h2><ul><li>Pink hair character again.</li><li>checkpoint is not properly known her so I trained this model.</li><li>all images is generated with WAI-IL (832x1216 + Hi-Res Fix 2x)</li></ul><h2 id="trigger-word">trigger word</h2>debut costume<pre><code>raora panthera, raora panthera \(1st costume\), panther girl, panther ears, animal ear fluff, blush, pink hair, streaked hair, large breasts, long hair, yellow eyes, goggles, black tube top, black hat, white thighhighs, cleavage, ascot, bare shoulders, thigh strap, white shorts, white vest, gloves, pink jacket, ankle boots, white boots, panther tail</code></pre>new year costume<pre><code>raora panthera, raora panthera \(new year\), japanese clothes, kimono, pink hair, white kimono, long hair, panther ears, yellow eyes, panther girl, large breasts, cleavage, multicolored hair, twintails, streaked hair, pink nails, nail polish, bare shoulders, white hair, thighhighs, white thighhighs, very long hair, animal ear fluff, off shoulder, wide sleeves, side slit, sidelocks, double-parted bangs, center-flap bangs, hair between eyes</code></pre>any costume<pre><code>raora panthera, panther girl, panther tail, animal ears, animal ear fluff, blush, pink hair, streaked hair, large breasts, long hair</code></pre>alternate hairstyle (replace hairstyle)<pre><code>ponytail</code></pre><h3 id="limitations">Limitations</h3><ul><li>jacket might not be remove properly (debut costume)</li></ul><h2 id="training-details">Training Details</h2>LoRA rank<ul><li>Lycoris LoCon</li></ul>dataset<ul><li>326 images debut costume</li><li>382 images others</li><li>67 images new year costume</li></ul>parameters<ul><li>resolution = 1024</li><li>batch size = 2</li><li>network dim,alpha = 16,16</li><li>conv dim,alpha = 16,16</li><li>mix/save precision = bf16/bf16</li><li>optmizer = Lion (weight_decay=0.1)</li><li>UNet LR = 4e-5</li><li>TE LR = 1.25e-05</li><li>scheduler = cosine_with_min_lr (min_lr_ratio 0.4)</li><li>l2 loss</li></ul>steps<ul><li>epochs = 10</li><li>total steps = 5540</li><li>repeat = 1,1,6</li></ul>tools<ul><li>kohya-ss GUI v24.3.0 FORKED by vjumpkung</li><li>torch 2.5.0 cu124</li><li>Runpod RTX 4090</li></ul>avg weight<ul><li>UNet average weights : 0.009681653650477528</li><li>TE1 average weights : 0.008030779737358293</li><li>TE2 average weights : 0.006554972336743958</li></ul>*This LoRA is for studying LoRA training with new technique so do not use for damaging the vtuber (also support her too).*If you see my model reposted on other image generation service please report to me.

Files with SHA256: d30d70904004750f40c3c584900552975d310fe9ff7ddbf4edd8a1665192b73b