<h2 id="25-06-16-updated-v2!-c3cxztdd1">25-06-16 UPDATED V2!</h2>A large amount of updated data based on in-game modeling was trained using ILXL 2.0. The outfit details have been significantly improved, especially with the Tacet mark on the feet now being represented.Major Changes<ol><li>The Tacet mark on the feet is now successfully implemented.</li><li>Outfit detail has been improved, and prompts have been refined. (Back view was also included in training.)</li><li>The model was retrained based on in-game modeling rather than original illustrations, resulting in facial details that more closely resemble the in-game appearance.</li><li>Due to the increased volume of training data, the style remains stable even at higher LoRA weights (0.8 and above), and it works well up to 1.0.</li></ol>Remaining Limitations<ul><li>The crescent-shaped bracelet is still not rendered properly. Even though a lot of training data was added, this part still needs improvement.</li></ul>This is a LoRA of the character Iuno from Wuthering Waves.Since I use ComfyUI and connect the LoRA through nodes, there are no specific trigger words required.If you're using a different tool instead of ComfyUI, you can use the <code>&lt;lora:~&gt;</code> syntax to call the LoRA.The base model used for training is ILXL, and the recommended LoRA weight is 0.8 to 1.0. Retrained as v2, it can now be used at weight 1.0 without any issues with art style or facial details!For the checkpoint used in the example images, I used the WaiNSFW model.Recommended prompt:<pre><code>1girl,solo,blue hair,gradient_eyes,blue_eyes,purple_eyes,twintails,
head_wreath,gold_ornament,wing_ornament,head wreath,black ribbon</code></pre>For the outfit, you can use:Top (upper body)<pre><code>large_breasts,gold ring,armlet,jewelry,
white dress,gold bra,o-ring_top,gold_wrist_cuffs,bracelet,gold collar,gold trim,tassel,bare_shoulders,single_sleeve_past_wrist</code></pre>Skirt (cowboy shot)<pre><code>blue showgirl skirt,pelvic_curtain,white_panties,single thigh strap,see-through_clothes</code></pre>Foot (full body)<pre><code>white sandals,tacet_mark_\(wuthering_waves\),feet,toes,foot_tattoo,anklet,</code></pre>주요 변경 사항 <ul><li>1. 발쪽의 Tacet mark가 성공적으로 구현됩니다. </li><li>2. 복장의 디테일 상승 및 프롬프트 정리를 했습니다. (뒷모습도 학습했습니다.) </li><li>3. 기존 일러스트 기반에서 인게임 모델링 기반으로 재학습하여 얼굴 디테일이 인게임 모델링에 유사하게 변경되었습니다. </li><li>4. 학습 데이터량이 많아져 0.8이상의 높은 Lora Weight에서도 그림체가 잘 유지되며 1.0까지 사용해도 무리 없습니다. </li></ul>아직 부족한 부분<ul><li> 반달모양 팔찌 구현이 여전히 안 됩니다. 학습 데이터를 많이 넣었는데도 이 부분은 아직 부족하네요.</li></ul> '워더링 웨이브'의 캐릭터 'Iuno'의 Lora입니다. 저는 comfyUI에서 노드로 연결하여 Lora를 호출하여 사용하기에 트리거워드는 따로 없습니다.만약 comfyUI가 아닌 다른 툴에서 사용 시에는 &lt;lora:~&gt; 하는 문법으로 호출하시면 사용 가능할 겁니다! 학습은 ILXL 베이스모델을 사용하였으며 추천하는 설정 값은 Lora weight 0.8~1.0 정도입니다.v2로 재학습하여 이제 1.0으로 사용해도 그림체 및 얼굴 디테일에 문제가 없습니다.그림을 뽑는 체크포인트의 경우 저의 경우엔 WaiNSFW 모델을 사용하였습니다. 추천하는 프롬프트는 다음과 같습니다.기본 외형<pre><code>1girl,solo,blue hair,gradient_eyes,blue_eyes,purple_eyes,twintails,
head_wreath,gold_ornament,wing_ornament,head wreath,black ribbon</code></pre>복장상의 (upper body)<pre><code>large_breasts,gold ring,armlet,jewelry,
white dress,gold bra,o-ring_top,gold_wrist_cuffs,bracelet,gold collar,gold trim,tassel,bare_shoulders,single_sleeve_past_wrist</code></pre>하의 (cowboy shot)<pre><code>blue showgirl skirt,pelvic_curtain,white_panties,single thigh strap,see-through_clothes</code></pre>발 (full body)<pre><code>white sandals,tacet_mark_\(wuthering_waves\),feet,toes,foot_tattoo,anklet,</code></pre>

Files with SHA256: f5188ac113b5b6ee5c6fa20ad118e687ae523abc0e4bc0392d76829b68c6ae82