<pre><code>**Tekka** is a photorealistic hybrid-face LoRA: realistic skin texture and photographic
backgrounds, with narrow, ukiyo-e-derived eyes and bone structure learned from
19th-century bijin-ga woodblock prints (public domain). It keeps the image fully
photographic — the print style only shapes the face. For the illustration-side versions
of this face family, see my sister LoRAs
[Inase](https://civarchive.com/models/2762597),
[Eejyanaika](https://civarchive.com/models/2765906) and
[the beretMix port](https://civarchive.com/models/2770385).

**How to use**
- Trained for **perfectionAsianILXL v1.0** (Illustrious-based realistic merge).
 Other models untested.
- Trigger: `soyokaze` / Weight: **0.8** (0.5 = subtle, closer to base face;
 **1.0 turns the whole image into a woodblock print — don't**)
- **Required negatives** (without these, kimono/traditional tags flip the image
 into print style): `ukiyo-e, woodblock print, traditional media, painting,
 illustration, sketch, flat color, sepia, monochrome`
- Full suggested negative: `bad quality, worst quality, blurry, lowres,
 (large eyes:1.4), (round eyes:1.4), big pupils, wide-eyed, eyelashes, eyeliner,
 eyeshadow, (double eyelid:1.3), 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, cross-eyed,
 strabismus, unfocused eyes`
- Positive anchors that help: `photo realistic, photography, realistic skin texture,
 depth of field, pale skin, black hair, black eyes`
- Want the eyes even narrower? Add boosters: `(narrow eyes:1.4), tsurime, monolid,
 single eyelid, small mouth`
- Suggested: Euler a, 28 steps, CFG 6. ADetailer (face_yolov8n, denoise 0.35)
 with the same LoRA at the same weight in the ADetailer prompt.
- Do NOT stack with other face/style LoRAs.

Trained exclusively on public-domain artworks — no AI-generated images, no copyrighted material.
</code></pre>

tekka

First release (epoch 2 of a 10-epoch run — chosen for the most natural photorealistic face).Trained on perfectionAsianILXL v1.0. Weight 0.8 recommended (0.5 = subtle).Do not go above 0.8 — at 1.0 the whole image flips into a woodblock-print style.Keep the style negatives from the description (ukiyo-e, woodblock print, painting, etc.)or traditional-clothing tags will trigger the same flip.

Files with SHA256: 48e6b7cab15a3a60f7378731023e197820a90e5bdea560be8f615d4506c50624