<h2 id="my-workflow">My Workflow</h2><a target="_blank" rel="ugc" href="https://gist.github.com/hinablue/7c35976d4d58f54f4016d3656034e77e">https://gist.github.com/hinablue/7c35976d4d58f54f4016d3656034e77e</a><hr /><h2 id="v16-v16-x-large-medium">v16, v16-X - Large / Medium</h2>使用多解析度訓練，並採用自行製作的優化器（<a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/README.md#hinaadaptive-%E8%A8%98%E6%86%B6%E9%AB%94%E5%84%AA%E5%8C%96%E8%87%AA%E9%81%A9%E6%87%89%E7%89%88%E6%9C%AC-">HinaAdaptive</a>）與 <a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/FOURIER_LOSS_GUIDE.md">FFT Loss Function</a>。Use multi-resolution training, and adopt a self-made optimizer (<a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/README.md#hinaadaptive-%E8%A8%98%E6%86%B6%E9%AB%94%E5%84%AA%E5%8C%96%E8%87%AA%E9%81%A9%E6%87%89%E7%89%88%E6%9C%AC-">HinaAdaptive</a>) and <a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/FOURIER_LOSS_GUIDE.md">FFT Loss Function</a>.マルチ解像度トレーニングを使用し、自作のオプティマイザー（<a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/README.md#hinaadaptive-%E8%A8%98%E6%86%B6%E9%AB%94%E5%84%AA%E5%8C%96%E8%87%AA%E9%81%A9%E6%87%89%E7%89%88%E6%9C%AC-">HinaAdaptive</a>）と<a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/FOURIER_LOSS_GUIDE.md">FFT損失関数</a>を採用します。다중 해상도 학습을 활용하고 자체 제작한 최적화기(<a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/README.md#hinaadaptive-%E8%A8%98%E6%86%B6%E9%AB%94%E5%84%AA%E5%8C%96%E8%87%AA%E9%81%A9%E6%87%89%E7%89%88%E6%9C%AC-">HinaAdaptive</a>)와 <a target="_blank" rel="ugc" href="https://github.com/hinablue/sd-scripts/blob/sd3-hina/docs/hina/FOURIER_LOSS_GUIDE.md">FFT 손실 함수를</a> 채택합니다.已知問題：<ol><li>肢體、手、腳的表現仍不佳。</li></ol>Known Issues:<ol><li>The performance of the limbs, hands and feet is still poor.</li></ol>既知の問題:<ol><li>手足のパフォーマンスは依然として低いです。</li></ol>알려진 문제:<ol><li>사지, 손, 발의 성능은 여전히 ​​좋지 않습니다.</li></ol><hr /><h2 id="v3v2-large-and-v4v3-medium">v3/v2 - Large &amp; v4/v3 - Medium</h2>已知問題：<ol><li>肢體、手、腳的表現仍不佳。</li><li>五官在小面積表現不佳。</li></ol>Known Issues:<ol><li>The performance of the limbs, hands and feet is still poor.</li><li>Facial features perform poorly in small areas.</li></ol>既知の問題:<ol><li>手足のパフォーマンスは依然として低いです。</li><li>顔の特徴は、狭い領域ではパフォーマンスが低くなります。</li></ol>알려진 문제:<ol><li>사지, 손, 발의 성능은 여전히 ​​좋지 않습니다.</li><li>얼굴 특징은 작은 영역에서 성능이 좋지 않습니다.</li></ol><hr /><h2 id="v2-rev10-medium">v2-rev10 - Medium</h2>這是使用分層、分塊與調整 shift 參數重新訓練 12 個相關 Lora 的整合版本，可以表現出真實皮膚與高度細節。This is an integrated version of 12 related Lora retrained using layering, chunking and adjusting the shift parameter to represent realistic skin and high detail.これは、リアルな肌と高いディテールを表現するために、レイヤー化、チャンキング、シフト パラメータの調整を使用して再トレーニングされた 12 個の関連する Lora の統合バージョンです。이는 레이어링, 청킹 및 이동 매개변수 조정을 사용하여 현실적인 피부와 높은 디테일을 표현하기 위해 재교육된 12개의 관련 Lora 의 통합 버전입니다.已知問題：<ol><li>肢體、手、腳的表現仍不佳。</li></ol>Known Issues:<ol><li>The performance of the limbs, hands and feet is still poor.</li></ol>既知の問題:<ol><li>手足のパフォーマンスは依然として低いです。</li></ol>알려진 문제:<ol><li>사지, 손, 발의 성능은 여전히 ​​좋지 않습니다.</li></ol><hr /><h2 id="v1-rev6-large">v1-rev6 - Large</h2>這是一個非常早期的測試。This is a very early version of the test.これはテストの非常に初期バージョンです。이것은 테스트의 아주 초기 버전입니다.已知問題：<ol><li>肢體、手、腳的表現仍不佳。</li><li>五官在小面積表現不佳。</li></ol>Known Issues:<ol><li>The performance of the limbs, hands and feet is still poor.</li><li>Facial features perform poorly in small areas.</li></ol>既知の問題:<ol><li>手足のパフォーマンスは依然として低いです。</li><li>顔の特徴は、狭い領域ではパフォーマンスが低くなります。</li></ol>알려진 문제:<ol><li>사지, 손, 발의 성능은 여전히 ​​좋지 않습니다.</li><li>얼굴 특징은 작은 영역에서 성능이 좋지 않습니다.</li></ol><hr /><h2 id="v1-rev2-medium">v1-rev2 - Medium</h2>這是一個非常早期的測試。This is a very early version of the test.これはテストの非常に初期バージョンです。이것은 테스트의 아주 초기 버전입니다.已知問題：<ol><li>臉型在部分角度呈現上會有左右臉失衡的狀況。</li><li>肢體、手、腳的表現仍不佳。</li><li>五官在小面積表現不佳。</li></ol>Known Issues:<ol><li>The face shape may appear imbalanced between the left and right faces at certain angles.</li><li>The performance of the limbs, hands and feet is still poor.</li><li>Facial features perform poorly in small areas.</li></ol>既知の問題:<ol><li>特定の角度では、顔の形が左右で不均衡に見える場合があります。</li><li>手足のパフォーマンスは依然として低いです。</li><li>顔の特徴は、狭い領域ではパフォーマンスが低くなります。</li></ol>알려진 문제:<ol><li>특정 각도에서 얼굴 모양이 좌우 얼굴 사이에서 불균형하게 보일 수 있습니다.</li><li>사지, 손, 발의 성능은 여전히 ​​좋지 않습니다.</li><li>얼굴 특징은 작은 영역에서 성능이 좋지 않습니다.</li></ol><hr />本模型所使用的訓練素材，皆以 <a target="_blank" rel="ugc" href="https://civarchive.com/models/176449/9527-detail-realistic-xl">9527 Detail Realistic XL </a>所產生之東方臉孔為主要訓練素材，其中也包含了我自行擁有之訓練素材。The training materials used in this model are all oriental faces generated by <a target="_blank" rel="ugc" href="https://civarchive.com/models/176449/9527-detail-realistic-xl">9527 Detail Realistic XL</a> as the main training materials, which also include training materials I own.このモデルで使用されているトレーニング マテリアルはすべて、<a target="_blank" rel="ugc" href="https://civarchive.com/models/176449/9527-detail-realistic-xl">9527 Detail Realistic XL</a> によって生成された東洋人の顔で、メインのトレーニング マテリアルとして使用されています。これには、私が所有するトレーニング マテリアルも含まれています。이 모델에 사용된 교육 자료는 모두 <a target="_blank" rel="ugc" href="https://civarchive.com/models/176449/9527-detail-realistic-xl">9527 Detail Realistic XL</a> 에서 생성한 동양인 얼굴이며, 주요 교육 자료에는 내가 소유한 교육 자료도 포함됩니다.

[SD3.5] Asian Mix Medium / Large Lora

hinaSD35AsianMedium_v16.safetensors

v16 - Medium

v16-X Medium

v16-X Large

v16 - Large

v4.0 - Medium

v3.0 - Large

v3.0 - Medium

v2.0 - Large

v2.0-rev10 - Medium

v1.0-rev6 Large

v1.0-rev2 - Medium

Files with SHA256: 162b6561173d1ceda1357834ea8196c11ef9ca16c3da71974f9af7d4f7b4765f