This is a character LoRA for the original character Yui Kohaku (琥珀ゆい), specifically trained for the Illustrious-XL model.Yui is characterized by her elegant long black hair, traditional hime cut (side bangs), and neat blunt bangs. She has a calm and charming appearance with brown eyes.<h3 id="recommended-settings">Recommended Settings</h3><ul><li>Trigger Word: <code>YuiKohaku</code></li><li>LoRA Weight: <code>0.8 - 1.0</code> (Start at 1.0 for high accuracy, or lower it slightly for more dynamic poses)</li><li>Base Model: Illustrious-XL</li><li>Character Tags: <code>1girl, solo, long hair, black hair, blunt bangs, brown eyes, hime cut</code></li></ul><h3 id="training-details">Training Details</h3><ul><li>Dataset: 55 high-quality hand-curated images.</li><li>Best Version: Epoch #8 was selected for providing the perfect balance between character fidelity and prompt flexibility.</li><li>Versatility: Thanks to tag shuffling and a diverse dataset, this LoRA works great with various outfits including school uniforms, casual hoodies, traditional kimonos, and more.</li></ul><hr /><h2>日本語説明</h2>Illustrious-XL専用に学習させた、オリジナルキャラクター「琥珀ゆい」のLoRAです。 黒髪ロング、姫カット、ぱっつん前髪が特徴的な美少女です。<ul><li>トリガーワード: <code>YuiKohaku</code></li><li>推奨強度: 0.8 〜 1.0</li><li>特徴: 55枚の高品質な画像で学習を行い、最も再現度と汎用性のバランスが良かった「Epoch #8」を採用しました。制服、パーカー、和装など、衣装の着せ替えがスムーズに行えるよう調整しています。</li></ul>

Yui Kohaku / 琥珀ゆい

Initial Release (v1.0)This is the first stable release of the Yui Kohaku LoRA. After a total of 15 epochs of training, I have selected Epoch #8 for this version as it provided the most consistent facial features while maintaining excellent flexibility for different outfits and poses.Key Features:<ul><li>Optimal Balance: Epoch 8 avoids the "pose locking" (overfitting) seen in later epochs, allowing for easier character manipulation.</li><li>High Fidelity: Captures Yui's signature hime cut and blunt bangs with high precision.</li><li>Outfit Flexibility: Successfully tested with school uniforms, casual hoodies, and traditional Japanese kimonos.</li></ul>日本語メモ: 琥珀ゆい LoRAの初版リリースです。 全15エポックの学習結果を比較検討し、最も「顔の再現度」と「ポーズ・衣装の柔軟性」のバランスが良かった Epoch 8 を採用しました。 後半のエポックで見られた「構図の固定化」が起きにくいため、様々なシチュエーションで使いやすいモデルになっています。

Files with SHA256: 48d52653c6d930d6156d3613005280f47bf21e4f56ab9858b37d7dee396e12e4