# seigan — serene classical beauty LoRAseigan(清顔) gives photorealistic female portraits a serene, classicallysculpted beauty: narrow almond eyes with a meditative, half-closed gaze (半眼),a calm expressionless mouth, and finely balanced features.- Base model: beretMixReal_v110 (Illustrious) — other realistic Illustrious/SDXL merges may work but are untested- Trigger word: <code>seigan</code>- Training data: a small curated set of freely-licensed images ## Recommended recipe- **Weight 0.6** for portraits / upper body. **Raise to 0.8 for face close-ups** (the effect fades in extreme close-ups at 0.6)- The booster <code>(narrow eyes:1.3), tsurime, single eyelid, black eyes</code> is part of the recipe — the LoRA is trained light so skin stays photographic- Euler a, 28 steps, CFG 6. Hires 2x (denoise 0.5–0.6) for quality- **Face close-ups: use a square canvas** (1024×1024 → hires 1.6x, denoise 0.5). Tall canvases can duplicate facial features at high resolution## Negative prompt — important!**Do not put <code>statue, sculpture, figurine, doll</code> in the negative prompt** —these words cancel this LoRA's effect entirely. Use texture words only:Optional extras:- Perfectly smooth skin: add <code>wrinkles, aged skin, nasolabial folds, eye bags</code> to NEG- Meditative half-closed eyes (半眼): add <code>expressionless, serene expression</code> to POS and <code>smile, grin, open mouth, teeth</code> to NEG---seigan(清顔)は、切れ長の半眼と端正な造形を実写ポートレートに与える LoRA です。トリガー <code>seigan</code>、推奨重み 0.6(顔アップは 0.8)+上記ブースター。NEG に statue / sculpture 等を入れると効果が消えます(質感語のみ使用)。学習素材はすべてフリーライセンス(CC0 / CC BY / CC BY-SA、全件ライセンス・作者記録済み)で、実在人物の画像は使用していません。

seigan

First release. Trained on beretMixReal_v110 (Illustrious), 20 curatedfreely-licensed images, 2240 steps / 14 epochs.Use weight 0.6 for portraits (0.8 for face close-ups) together with the boostertags — see the model description for the full recipe and the negative-promptwarning (never put "statue / sculpture" in NEG).初版。推奨重み 0.6(顔アップ 0.8)+ブースタータグ併用。NEG に statue / sculpture を入れると効果が消えるので注意(詳細は説明文)。

Files with SHA256: 580bb40e6d085f7b32b6f1a9b9efd5ce3066ac1cfee9406dd1dc2f5cef2bb8cf