Quedlinburg — Architectural Concept & Materiality LoRA

Trained on Krea-2 using a curated dataset of 55 high-resolution, uncompressed DSLR photographs, this LoRA brings the authentic architectural soul and physical textures of Quedlinburg, Germany—a UNESCO World Heritage Site—directly into your generations.

While designed as a location model for medieval half-timbered architecture (*Fachwerk*), cobblestone alleys, and Romanesque stone masonry, this LoRA is built to be a universal background and material engine. It doesn’t just render a town—it fundamentally upgrades how the base model handles light, stone, and physical surfaces.

What this LoRA does for your generations:

* Eliminates the "AI Plastic Sheen": This LoRA injects tangible real-world physics into your scenes—porous sandstone blocks, hand-hewn oak timber grain, weathered mortar, and authentic tactile stone relief.

* Flawless Character & Fashion Backdrops: Designed with strict concept disentanglement. You can place modern fashion models, travel influencers, or fantasy characters into the scene without worrying about "character corruption"—skin textures, makeup, and contemporary clothing (like knitwear or trench coats) remain pristine and untouched.

* Complete Style Freedom (Zero Style Lock-In): This is not locked to photography! Because the dataset was cleanly captioned, the architectural DNA transfers effortlessly into 2D Studio Ghibli anime, 19th-century oil paintings, or high-concept sci-fi.

* Mastery of Seasons & Lighting: Tested and proven across all atmospheric conditions: from bright summer ice-cream café patios to rain-slicked night streets, volumetric church sunbeams, and snowy twilight Christmas markets (*Advent in den Höfen*).

* High-Tech & Automotive Synergy: The model effortlessly blends historic architecture with modern and sci-fi elements (supercars, hovercrafts, neon signage) with accurate ray-traced reflections on wet cobblestones.

Recommended Settings & Usage

* Activation / Trigger: qdlbg

* LoRA Weight / Strength: 0.85 – 1.0 (Use 1.0 for full architectural accuracy; dial down to 0.8 for softer ambient backgrounds).

* Target Architecture: Krea-2 / DiT models (optimized for natural language prompting).

* Passive Isolation: When the trigger word is not used, this LoRA remains completely neutral and will not contaminate unrelated prompts (like modern skyscrapers or beaches).

### Prompting Tips:

* For Historic Realism: Mention specific details like half-timbered houses, dark oak beams, steep red clay roofs, blooming window flower boxes, and wet cobblestones.

* For Portraits: Pair with an 85mm or 50mm portrait lens prompt at f/1.8 for creamy background bokeh that turns the town into an aesthetic European backdrop.

* For Genre Bending: Don't hesitate to prompt 2D anime illustration, impressionist oil painting, or cyberpunk neon sci-fi—the LoRA adapts its architecture to whatever medium you specify!