Authentic Qassim Klega: The Ultimate SDXL LoRA Dataset & Guide 🍪✨

Welcome to the official documentation and showcase for the Qassim_Klega LoRA, specifically trained for SDXL. This specialized model brings the rich culinary heritage of Najd "QASSIM"into the world of AI generation with stunning, photorealistic accuracy.

1. Introduction: What is Qassim Klega? 🇸🇦

Klega (الكليجا) is the crown jewel of traditional Saudi hospitality, deeply rooted in the historic Al-Qassim region. It is a golden, hollow-centered pastry famous for its iconic crisp geometric grid pattern. Inside its hard, dome-shaped crust lies a rich, steaming filling made from date molasses, dark lime (loomi), cinnamon, and cardamom. Generating this unique pastry accurately in AI has always been a challenge due to its highly specific exterior texture and the viscous look of its molasses filling—until now.

2. A Hyper-Niche Architectural Masterpiece 🎯

This LoRA targets a hyper-niche community of digital artists, commercial designers, and cultural curators. Instead of generalized baked goods, this model is hyper-focused on capturing the true, authentic identity of Klega. Whether you are generating marketing materials for premium retail packaging or historical Saudi culinary setups, this model provides the precise textures and cultural context required for professional-grade output.

3. Training Parameters & Configuration 🛠️

To achieve maximum fidelity, the LoRA was trained using the exact production parameters optimized for SDXL's demanding architecture. The precise setup utilized is outlined below:

LoRA Type: Standard (SDXL)
Resolution: -1024 \times 1024 pixels
Epochs: 10
Num Repeats: 14
Batch Size: 1
Unet Learning Rate: 0.0001
Text Encoder Learning Rate: 0.00005
LR Scheduler: constant_with_warmup (with 1 cycle)
Min SNR Gamma: 5 (For contrast stability and deep tones)
Optimizer: Adafactor
Network Dim / Alpha: 64 / 32
Noise Offset: 0.05 (Essential for capturing the deep, glossy shades of dark syrup)
Data Augmentation: Enable Bucket, Shuffle Tags, Flip Augmentation enabled.
Keep Tokens: 1 (Secures the trigger word at the absolute front)

4. The Magic of Clip Skip 2 🧠

For this SDXL training, Clip Skip is set to 2.

Unlike older base models, skipping the final layer allows the network to tap into the conceptual and stylistic layer of the text encoder. This ensures the model flawlessly balances the rigid geometric squares of the pastry with flexible prompt instructions, preventing the model from becoming too stubborn when you change backgrounds or lighting conditions.

5. Dataset Curation & Enhancements 📸

The underlying training dataset is comprised of 18 pristine, ultra-high-definition images (1024 \times 1024).

Every single image underwent strict preprocessing to enhance native sharpness and correct color profiles to match real-life daylight and studio conditions. The dataset spans an extensive range of styles—from single structural close-ups and dramatic, slow-pouring molasses macro shots to contemporary retail presentations, including traditional iron tins and modern, transparent plastic packaging.

6. Suggested Prompts (8 Creative Variations) 📝

Always start your prompts with the official trigger word: Qassim_Klega. Here are 8 tailored prompt configurations using our exact tag structures:

1. The Classic Pile (White Background)

Qassim_Klega, 5pcs, stacked, sesame seeds topping, whole, round shape, puffed dome, geometric grid pattern, golden brown crust, baked texture, close-up shot, clean off-white background, studio lighting.

2. The Traditional Heritage Setting

Qassim_Klega, multiple, stack, piling up, in a traditional woven palm-frond basket, colorful woolen tassels, traditional saudi decorations, golden brown crust, detailed grid pattern, glossy finish, side view, festive context, soft indoor lighting.

3. The Dramatic Spill (Macro)

Qassim_Klega, multiple, broken open, heavy date molasses flow, thick dark syrup spilling onto surface, glossy liquid, wet texture, pieces of crispy baked crust, detailed grid pattern, close-up, rustic wooden table surface, dramatic side lighting, high contrast.

4. The Commercial Studio Spread

Qassim_Klega, 3pcs, arranged in a row, whole, side view, distinct puffed dome shape, thick crust, uniform geometric grid pattern, golden brown color, baked, solid white background, commercial product photography.

5. Individual Premium Wrapped (Retail)

Qassim_Klega, 6pcs, mini Kleicha, small size, individually wrapped, clear plastic packaging, transparent wrappers, sealed bags, arranged on a surface, golden brown crust, grid pattern, commercial product display, indoor lighting.

6. The Royal Platter

Qassim_Klega, multiple, stack, cut open, date molasses filling pouring down, glossy dark liquid, golden brown crust, crisp grid pattern, arranged on an ornate golden metal platter, colorful jewel embellishments, isolated on a solid white background, studio commercial photography.

7. The Cross-Section Reveal

Qassim_Klega, 1pc, broken open, cut in half, cross-section view, revealing hollow inside, dark date molasses filling layer, dry spice coating, thick baked dough crust, detailed grid pattern on top, rustic style, simple background.

8. The Vintage Iron Tin (Unboxing)

Qassim_Klega, multiple, stacked, piled inside, open traditional metal tin container, round iron box, packaging, vintage metallic canister, numerous whole biscuits, golden brown crust, geometric grid pattern, product display, top-down angle view, indoor lighting.

7. Recommended KSampler Settings ⚙️

To achieve optimal inference rendering with this LoRA in your favorite WebUI or ComfyUI workflow, use these baseline sampler settings:

Base Model: Premium SDXL Photorealistic Checkpoints (e.g., Realistic Stock Photo, Juggernaut XL, or Fooocus native).
LoRA Weight: 0.85 to 1.10 (0.93 is the sweet spot).
Sampling Steps: 30 to 35 steps.
Sampler (KSampler): DPM++ 2M SDE
Scheduler: Karras or SGM Uniform.
CFG Scale: 5.0 to 6.5 (Keep it moderately low for natural, non-burned skin/crust textures).

8. Conclusion & Support ☕🌟

This model represents a deep passion for digital preservation and regional culinary art. Bringing highly detailed textures like Klega to the open-source community takes immense time, meticulous data curation, and testing.

If this model elevates your workflow, adds value to your retail concepts, or simply brings a taste of Saudi heritage to your digital canvas, please consider showing your support!

Drop a 5-star review, share your generations in the community gallery, Your engagement keeps these specialized projects alive. 🚀✨

FASIAL ..