<h2 id="description">Description</h2>This LoRA is designed for WAI-Nsfw-Illustrious-17 and related Illustrious XL anime checkpoints. It targets Xunlan's Gongpinglan outfit with emphasis on identity consistency, clean facial recall, rabbit ears, hair accessories, tassel earrings, and the red-black formal outfit silhouette.The recommended prompt keeps both the unique trigger and light character anchors:<pre><code>xunlan_gongpinglan, xunlan, arknights, 1girl</code></pre>The model was selected after an epoch and weight evaluation matrix. Epoch 8 at weight 0.8 gave the best balance between identity strength, outfit recall, and motion flexibility. Higher weights can strengthen the outfit but may reduce pose freedom.No NSFW images were included in the training dataset. The showcase set is SFW / suggestive and was regenerated with the recommended LoRA rather than copied from the training set or evaluation matrix.<h2 id="recommended-usage">Recommended Usage</h2><h3 id="default-weight">Default Weight</h3><code>0.8</code><h3 id="useful-range">Useful Range</h3><code>0.7-0.9</code><h3 id="motion-composition-first">Motion / Composition First</h3>Use <code>0.7</code> when the pose, camera angle, or action is more important than strict outfit locking. This is useful for walking shots, seated poses, over-shoulder views, and scenes with stronger composition prompts.<h3 id="strong-identity-outfit-recall">Strong Identity / Outfit Recall</h3>Use <code>0.9</code> when the character identity or outfit details need more strength, especially for simple portraits or clean full-body references.<h3 id="avoid-as-default">Avoid as Default</h3>Avoid <code>1.0</code> as the default weight. It can make the outfit stronger, but it may also make poses more rigid and reduce prompt following.<h2 id="basic-prompt">Basic Prompt</h2><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.8&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, no text, looking at viewer, upper body, simple background, masterpiece, best quality, amazing quality, very aesthetic</code></pre><h2 id="full-character-prompt">Full Character Prompt</h2><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.8&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, pale blonde hair, straight hair, blunt bangs, short bob with long sidelocks, long hair behind shoulders, green hair ornament, long rabbit ears, pink eyes, expressionless, turquoise tassel earrings, black oversized blazer, red chinese-style crop top, high collar, chest cutout, black straps, round pendant, gold buttons, waist belt, white waist pouch, black fingerless gloves, wide-leg black pants, black ankle socks, black pointed-toe loafers, low heels, no text, masterpiece, best quality, amazing quality, very aesthetic</code></pre><h2 id="negative-prompt">Negative Prompt</h2><pre><code>low quality, worst quality, lowres, bad anatomy, bad hands, bad fingers, extra fingers, missing fingers, fused fingers, extra limbs, missing limbs, disfigured, deformed, blurry, jpeg artifacts, text, watermark, logo, signature, arknights logo, chinese text, english text, caption, artist name, bad face, bad eyes, multiple girls, 2girls, cropped head, cropped feet, out of frame, wrong hair color, dark hair, red eyes, blue eyes, missing rabbit ears, missing long rabbit ears, wrong outfit, missing black oversized blazer, missing red chinese-style crop top, missing waist pouch, missing wide-leg black pants, wrong footwear</code></pre><h2 id="example-prompts">Example Prompts</h2><h3 id="portrait">Portrait</h3><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.8&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, no text, no logo, no watermark, close-up portrait, beautiful face, detailed pink eyes, finger near lips, soft focus background, clear green hair ornament, clear turquoise tassel earrings, masterpiece, best quality, amazing quality, very aesthetic, absurdres</code></pre><h3 id="full-body">Full Body</h3><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.8&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, no text, no logo, no watermark, full body, contrapposto, one hand on hip, elegant pose, clean white studio background, clear black oversized blazer, clear red chinese-style crop top, clear waist pouch, clear wide-leg black pants, clear black pointed-toe loafers, masterpiece, best quality, amazing quality, very aesthetic, absurdres</code></pre><h3 id="seated-sensual">Seated / Sensual</h3><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.8&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, no text, no logo, no watermark, sitting on ornate chair, legs crossed, hand near face, elegant lounge, warm light, refined composition, clear pale blonde hair, clear rabbit ears, tasteful sensual mood, masterpiece, best quality, amazing quality, very aesthetic, absurdres</code></pre><h3 id="dynamic">Dynamic</h3><pre><code>&lt;lora:xunlan_gongpinglan_recommended_e000008:0.7&gt;, xunlan_gongpinglan, xunlan, arknights, 1girl, solo, no text, no logo, no watermark, full body, walking through modern city walkway at night, looking back, windblown pale blonde hair, soft neon rim light, clear silhouette, detailed outfit, masterpiece, best quality, amazing quality, very aesthetic, absurdres</code></pre><h2 id="recommended-generation-settings">Recommended Generation Settings</h2><ul><li>Checkpoint: <code>WAI-Nsfw-Illustrious-17.safetensors</code></li><li>Resolution: <code>1024x1024</code></li><li>Sampler: <code>DPM++ 2M</code> for showcase images, <code>Euler a</code> also works for quick testing</li><li>Steps: <code>28-30</code></li><li>CFG scale: <code>6.0-6.5</code></li><li>LoRA weight: <code>0.8</code> default, <code>0.7-0.9</code> useful range</li><li>Hires fix: optional; not required for the included showcase set</li></ul><h2 id="known-limitations">Known Limitations</h2><ul><li>The model is tuned for WAI / Illustrious XL style checkpoints. Pony, realistic XL, and non-Illustrious checkpoints may not preserve the outfit consistently.</li><li>High LoRA weight can reduce pose flexibility and may make the composition more rigid.</li><li>Low LoRA weight can improve motion and camera freedom, but may weaken the red top, waist pouch, or blazer details.</li><li>Complex seated poses can trade off outfit accuracy against pose obedience.</li><li>Stacking with strong style LoRAs or other outfit LoRAs may cause clothing conflicts.</li><li>Back-view and strict over-shoulder shots may need explicit pose wording and a lower LoRA weight.</li></ul><h2 id="training-summary">Training Summary</h2><ul><li>Training images: <code>100</code></li><li>Caption files: <code>100</code></li><li>SFW samples: <code>100</code></li><li>NSFW samples: <code>0</code></li><li>Base model: <code>WAI-Nsfw-Illustrious-17.safetensors</code></li><li>Training type: SDXL LoRA, UNet-only</li><li>Network dim: <code>32</code></li><li>Network alpha: <code>16</code></li><li>Network dropout: <code>0.05</code></li><li>Optimizer: <code>AdamW8bit</code></li><li>Precision: <code>bf16</code></li><li>Epochs: <code>10</code></li><li>Recommended epoch: <code>8</code></li><li>Final training loss reference: about <code>0.0651</code></li><li>Recommended model file: <code>xunlan_gongpinglan_recommended_e000008.safetensors</code></li></ul>

Files with SHA256: 443b2ee980d5dfc067b1a4989da522149a2d655a28b4da6ac3ec13399bf95f9f