Downloads · 30 days
0
SHIFUKU-Gold-Leaf-SDXL-LoRA
TextureLoRALab/SHIFUKU-Gold-Leaf-SDXL-LoRA
SHIFUKU-Gold-Leaf-SDXL-LoRA is a machine learning model from TextureLoRALab. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as creativeml-openrail-m.
A texture LoRA trained on real gold leaf (金箔), not AI-generated images.
Access
Public
Updated Apr 5, 2026
Repo size
—
Likes
0
Public
Repo makeup
Click a slice to open those files.
.md3.3 KB · 68%
At a glance
- License
- creativeml-openrail-m
- Access
- Public
- Created
- Mar 1, 2026
- Updated
- Apr 5, 2026
- SHA
- 5f409088
Base models
Model card
From the Hugging Face model README
SHIFUKU Gold Leaf — Traditional Japanese Gold Leaf Texture LoRA
A texture LoRA trained on real gold leaf (金箔), not AI-generated images.
What makes this different
Most texture/style LoRAs are trained on AI-generated or digital images. This LoRA is trained on macro photographs of physical gold leaf — capturing grain structure, light diffraction, aging patina, and the subtle wrinkles left by craftsmen hammering the foil.
The training data comes from the real material, not from pictures of the material.
This is Texture Transplant (質感移植) — extracting texture from its original substrate and making it function in a new environment.
Training Details
| Parameter | Value |
|---|---|
| Base model | DreamShaperXL alpha2 (SDXL) |
| Training tool | kohya-ss/sd-scripts |
| Training images | 50 macro photographs of real gold leaf |
| Resolution | 1024×1024 |
| Epochs | 10 |
| Network dim / alpha | 128 / 64 |
| Learning rate | 1e-4 |
| Optimizer | AdamW8bit |
| Scheduler | cosine_with_restarts |
| Noise offset | 0.1 |
| Caption method | BLIP2 + manual correction |
Recommended Usage
- Trigger words:
gold leaf texture,kinpaku - Weight: 0.7–0.9
- Sampler: DPM++ 2M Karras, 25–35 steps
- Compatible with: SDXL 1.0 based models
Links
- Civitai: SHIFUKU Gold Leaf
- Blog: texture-lora-lab.com
- X: @TextureLoRALab
- Book: 「引き算のAI」(forthcoming)
About the creator
Fine art graduate (Japanese painting, Kanazawa College of Art) with an MA in Museum Studies (UK, Merit). Researching texture digitization through AI — bridging traditional craft materials and generative models.
Statement on artistic expression
I have no intention of infringing upon the expression of artists and craftspeople working outside the digital realm.
What I do — transplanting texture into AI — is an experiment. The generated images are countless sketches, not finished works.
This does not replace the hand of the craftsman who beats gold leaf, nor does it shortcut the layering of urushi lacquer. I do not pretend to reproduce the hamon a swordsmith forges into steel.
Texture Transplant is merely one means of recording the beauty held within physical materials and experimenting with it in digital space. The beauty of material itself exists only within the material.
License
Commercial use permitted. Credit appreciated but not required.
#SubtractiveAI #TextureTransplant #引き算のAI
🔗 研究サイト・技術ブログ
texture-lora-lab.com — TextureLoRALabの公式サイト
- 質感LoRA学習の完全マスターコース → texture-lora-lab.com/pack
- LoRA学習技術ブログ → texture-lora-lab.com/blog
- RunPod・Kohya_ssを使った学習ノウハウを公開中
- License
- creativeml-openrail-m
- Languages
- en, ja
- Created
- Mar 1, 2026
- Updated
- Apr 5, 2026