What is this?
A LoRA that captures the physical textures of a real Japanese sword (日本刀 / nihontō) — not a fantasy weapon, not a game asset, but the actual material surfaces: the hamon (刃文) temper line that emerges from differential hardening, the mirror-polish of the blade (地鉄 / jigane), the lacquered scabbard (鞘 / saya), and the gold-fitted guard (鍔 / tsuba).
This is not a weapon LoRA. This is a matière LoRA — it teaches the model what centuries-old steelwork and fittings feel like as physical surfaces.
⚠️ This is a BETA release. The dataset is still being refined. Feedback welcome — especially on hamon pattern accuracy and blade reflection behavior.
The Problem This Solves
AI-generated swords are almost always wrong. They look like plastic props or video game renders:
Blades are uniformly shiny with no internal structure
Hamon patterns (the wavy temper line) are missing or drawn as a simple wavy line
Scabbards look like painted tubes instead of lacquerware
Fittings (tsuba, habaki, menuki) lack metallic depth and craftsmanship detail
A real Japanese sword has at least 4 distinct material surfaces in a single object — each with completely different physical properties. This LoRA teaches the model to differentiate them.
How It Was Trained
Reference images of museum-quality Japanese swords, photographed at 3 distances:
Extreme close-up: Hamon crystal structure, steel grain (hada), gold inlay detail
Mid-range: Full blade with hamon line, scabbard lacquer sheen, fitting transitions
Full view: Complete sword in context, proportions, light interaction across surfaces
Trained with Kohya_ss on SDXL. Beta version — dataset will be expanded in future updates.
Recommended Setup
Base model: DreamShaper XL alpha2 (tested)
Weight: 0.5–0.7 for subtle material texture, 0.8–1.0 for full effect
Best subjects: Sword blades, katana compositions, weapon still life, samurai equipment
Trigger words: japanese sword, katana, hamon, nihonto, blade texture
Who Made This
TextureLoRALab — I studied Japanese painting (nihonga) in art school, earned a Master's degree in Museum Studies (with Merit) from a UK university, and now I combine traditional material knowledge with AI training techniques.
Japanese swords are some of the most complex material objects in art history — every surface tells a story about the smith's technique. I've studied these objects in museum collections, and that knowledge shaped the dataset.
Commercial Use
✅ Free for commercial use. No credit required.
More Textures in the SHIFUKU Series
🟡 Gold Leaf (金箔) — available
⚫ Kintsugi (金継ぎ) — available
🗡️ Hamon Steel (刃文) — this model (beta)
🟡 Gold Leaf v2 (金箔v2) — available
Follow TextureLoRALab on CivitAI to get notified when updates drop.
"Style LoRAs teach AI how things look. Matière LoRAs teach AI how things feel."
Description
v0.9 Beta — Initial release
Trained on museum-quality sword references using the 3-Distance Method
Covers: hamon temper line, jigane blade surface, saya lacquer, tsuba gold fittings
Known limitations: hamon pattern variety limited, works best with close-up compositions
Feedback welcome — DM on X or comment on CivitAI
Planned for v1.0: Expanded hamon variety (choji, midare, suguha), more tsuba designs, improved full-length sword compositions