nanites-mokuhanga-sdxl-1.0

SDXL LoRA generating images in the idiom of classic Japanese woodblock prints, Edo through late Meiji (c. 1670s–1910s): nishiki-e color prints with key-block line work, flat color fields, bokashi gradation, and period subject matter.

Trained on 199 curated public-domain / CC0 / CC-BY prints (bijinga, yakusha-e, fukei-ga, musha-e, kacho-ga). Corpus is private; rights log kept in-repo (SOURCES.md + manifest.csv).

"Ukiyo-e" means "pictures of the floating world." Range is Edo through late Meiji, not all of Japanese art. The corpus mixes period and reprint impressions (reprints tagged modern reprint); no first-edition provenance is claimed.

Trigger

Use mokuhanga in your prompt.

Example

mokuhanga, bijinga beauty with umbrella in snow, black key-block outlines, flat color fields, bokashi sky gradation, paper texture, cartouche with Japanese text, oban portrait

Negative

photorealistic shading, 3d render, anime, manga, modern objects, western single-point perspective, digital-clean vector, text, watermark, signature, frame, blurry, lowres

Training

base_model: stabilityai/stable-diffusion-xl-base-1.0
steps: 1500
lora_rank: 16
resolution: 768
lr: 0.0001
dataset_size: 199
loss_avg_last_50: 0.142
hardware: RTX 4070 8GB, CPU offload

Limitations

• Cartouches/seals garble: plausible print furniture, not legible text.
• Square-768 training center-crops oban ratios; inference at oban sizes.
• Style transfer only: blends motifs, no provenance or iconographic authority.
• Excludes shunga, Yokohama-e, surimono, shin-hanga/sosaku-hanga.

License

Adapter inherits CreativeML Open RAIL++-M (SDXL base). Reference images retain original licenses (PD/CC0/CC-BY); CC-BY attributions (e.g. AIC description text if reused) belong here.