A respectful homage to classic Japanese woodblock printmaking. This LoRA aims to reproduce crisp linework, a limited indigo/vermilion/sumi palette, washi paper grain, and subtle bokashi gradients from the Edo–Meiji tradition—without modern plastic lighting or heavy 3D shading. It’s built “the old way”: patient curation and careful training to keep compositions calm, balanced, and timeless.<h2 id="what-it's-good-at">What it’s good at</h2><ul><li>Portraits, coastal seascapes, lantern-lit streets, seasonal motifs (maple leaves, waves, mist).</li><li>Elegant kimono patterns, cartouches, publisher seals, and disciplined negative space.</li></ul><h2 id="training-summary-(civitai-lora-trainer-flux.1d)">Training summary (Civitai Lora Trainer / FLUX.1D)</h2><ul><li>Base model: FLUX.1-dev</li><li>Dataset: ~1000 curated ukiyo-e images, 1:1 @ 1024px</li><li>Epochs: 12 · Batch: 2 · Dim/Alpha: 16/16</li><li>UNet LR: 1e-4 · Text LR: 1e-5</li><li>Bucket: OFF (dataset is square) · Flip aug: OFF · Clip Skip: 1</li></ul><h2 id="usage">Usage</h2><ul><li>Works without a trigger. Optional helper tag: <code>ukiyoe_style</code>.</li><li>Add 2–3 style hints in the prompt, e.g. <code>ukiyo-e woodblock print, Japanese printmaking, washi paper texture</code>.</li><li>LoRA weight: 0.7–0.9 (push to 1.0 for a stronger print effect).</li></ul><h2 id="recommended-inference">Recommended inference</h2><ul><li>Resolution: 1024×1024 (or tasteful verticals 1024×1536 / 1024×2048)</li><li>Steps: 28–36 · CFG: 3.5–5.0</li><li>Sampler: DPM++ 2M Karras (or a comparable FLUX-friendly sampler)</li></ul><h2 id="negative-prompt-(short)">Negative prompt (short)</h2><code>modern logos, watermark, photoreal plastic look, harsh bloom, heavy 3D shading, text artifacts, extra fingers, gore</code>Tip: If the style “overprints,” lower LoRA weight to 0.6–0.7 or reduce CFG. If it feels too light, raise weight to 1.0 and emphasize <code>ukiyo-e woodblock print</code> in the prompt.

Files with SHA256: 7d79caec7b24c84bd54458b106e2d8ad8e390df05b7e40e59c59077b4a10cf18