[z-image] Realistic Cane Spanking Welts

📖 Introduction

This LoRA is specifically designed for the z-image model. It aims to accurately replicate the physiological textures of 3D, parallel, and swollen welts on the skin resulting from cane-based disciplinary actions.

---

⚙️ Training Specifications

• Dataset: 55 high-quality real-texture images.

• Steps: 16,500 Steps.

• Base Model: z-image-turbo-bf16.

---

💡 Weight & Usage Guide

To achieve the best results, please strictly follow these weight recommendations:

• Optimal Range: 0.2 - 0.5

• Warning Range: 0.5 - 0.8

  • Feature Shift: When the weight exceeds 0.5, the model will exhibit significant feature leakage. The character's facial features will irreversibly shift toward Western looks (even strong Asian-themed prompts may not be able to counteract this).

  • Structural Risks: When the weight approaches 0.8, the probability of structural artifacts (especially in hands and feet) increases significantly.

---

✍️ Prompting Strategy — CORE FOCUS!

This model is highly dependent on Natural Language Descriptions. If you only use vague terms like cane marks or red welts, the LoRA's effect will not be fully activated, and the textures may appear flat and uninspiring.

【Correct Usage】：

You must act like a "sculptor," providing detailed descriptions of the style, shape, and distribution of the marks.

Tip: It is strongly recommended to refer to the prompts provided in my Example Images.

---

💬 Author's Note

Since texture-based LoRAs perform differently across various lighting conditions and scenarios, I would greatly appreciate it if you could share your generated images and provide feedback!