<pre><code># Inase Real — Photorealistic × Ukiyo-e Face Hybrid

A photorealistic spin-off of **Inase** (https://civarchive.com/models/2762597), the
authentic ukiyo-e bijin-ga face LoRA from the Heiyamato series
(https://civarchive.com/models/2760524).

While the original Inase turns the whole image into an Edo-period woodblock
print, **Inase Real keeps the photograph and changes only the face grammar**:
realistic skin, real lighting, real backgrounds — but the elongated oval face
(urizanegao), narrow slit eyes, small mouth and nihongami hairstyle of
classical bijin-ga. A woman who stepped out of an old print into a modern photo.

Trained exclusively on public-domain artworks — no AI-generated images, no
copyrighted material. Same dataset as Inase, re-baked for a photorealistic base.

## How to use

- **Trigger**: `inase`
- **Base model**: beret mix Real v11.0 (trained and tested on it — do not use
  on other checkpoints)
- **Weight 0.5** (0.4 = subtle, 0.6 = stronger classical face; **0.8+ collapses
  the whole image into a woodblock print — don't**)
- Euler a / 28 steps / CFG 6
- Prompt template:
  ```
  masterpiece, best quality, photo realistic, photography, realistic skin texture, depth of field, inase, 1girl, solo, pale skin, &lt;scene tags&gt;, &lt;lora:inase_beret:0.5&gt;
  ```
- Negative (IMPORTANT — keep the style words, see Tips):
  ```
  bad quality, worst quality, blurry, lowres, large eyes, round eyes, big pupils, wide-eyed, eyelashes, eyeliner, eyeshadow, double eyelid, detailed eyes, makeup, 1boy, white skin, grey skin, blue skin, colored skin, 2girls, multiple girls, ukiyo-e, woodblock print, traditional media, painting, illustration, sketch, flat color, sepia, monochrome
  ```
- ADetailer (recommended): face_yolov8n, denoise 0.35, prompt
  `inase, 1girl, face, narrow eyes, small pupils, pale skin, small mouth, &lt;lora:inase_beret:0.5&gt;`

## Tips

- **Do not remove the style negatives** (`ukiyo-e, woodblock print, painting,
  illustration...`). Kimono and other traditional-clothing tags resonate with
  the LoRA and will drag the whole image into a print without them. With the
  negatives in place, kimono scenes stay fully photographic.
- The photo anchors in the positive prompt (`photography, realistic skin
  texture, depth of field`) matter for the same reason.
- The narrow eyes are baked into the trigger — do not prompt `closed eyes`.
- Use `pale skin`, not `white skin`.
- Works in modern scenes too (street, casual clothes) — the face and nihongami
  follow, the world stays contemporary.

This LoRA is a derivative of **beret mix Real** by AI_beret
(https://civarchive.com/models/1360673) — thank you for the excellent base model!

---

## 日本語

浮世絵美人画の「顔の文法」LoRA **Inase** の写実版スピンオフです。本家は画面
全体が版画になりますが、こちらは**写真のリアルさ(肌・光・背景)を保ったまま、
顔立ちだけを浮世絵型**(瓜実顔・切れ長の目・小さな口・日本髷)に寄せます。
学習データは著作権問題のないパブリックドメイン作品のみ(本家 Inase と同一
データセットを写実ベースで焼き直し)。

- トリガー: `inase` / 推奨重み **0.5**(0.4=控えめ、0.6=強め。0.8以上は全面
  版画化するので不可)
- ベース: beret mix Real v11.0 専用
- **ネガティブの様式語(ukiyo-e, woodblock print, painting 等)は削らないで
  ください。** 着物タグと共鳴して画面が版画化するのを防いでいます。
- プロンプト・設定の詳細は上の英語欄と同じです。
</code></pre><p></p>

inase_beret.safetensors

<p>First release. Epoch 8 of 10, adopted by visual sheet comparison. Keep the style negatives from the description — they prevent kimono scenes from collapsing into a woodblock print.</p>

Files with SHA256: 6f846ff4e1598f463eeb23bdd0430966b7700077d012214cc5948a983a20ec03