Inase — the ukiyo-e beauty (*bijin-ga*) face grammar of the Edo masters, now on Krea 2.

While most ukiyo-e LoRAs give you woodblock textures with modern anime faces, Inase learns the face grammar of classic Edo-period beauty prints and nothing else: the elongated oval face (*urizanegao*), narrow slit eyes, tiny mouth, and nihongami hairstyle, exactly as drawn in the golden age of bijin-ga.

One LoRA, two looks. This port takes advantage of Krea 2's photorealistic base:

- Weight 0.8–1.0 — full ukiyo-e stylization (the classic Inase look)

- Weight 0.6 — a realism hybrid: photoreal rendering with the Inase face grammar and mood

### How to use

- Trigger: inase

- Base model: Krea 2 (trained on RAW bf16, tested on Krea 2 Turbo nvfp4)

- Steps 8 (Turbo default) to 20 (more detail) / CFG 1.0 / euler + simple

- No negative prompt needed at CFG 1.0

- Prompt template:

masterpiece, best quality, inase, 1girl, solo, japanese clothes, kimono, pale skin, <scene tags>

- The narrow ukiyo-e eyes are baked into the trigger — do not fight them with open eyes.

- Krea 2 follows scene tags far better than SDXL: sci-fi environments, motorcycles, laboratories, and neon cityscapes all work while the face stays in style (see gallery).

Dataset: trained entirely on copyright-free material — a hand-curated set of 96 images selected from ~1,500 public-domain prints. No AI-generated images, no modern reproductions.

Training: musubi-tuner, dim 32 / alpha 32, lr 1e-4, 2,880 steps on krea2_raw_bf16. Epoch 4 of 6 selected by contact-sheet review.