Manga Rendering Style LoRA (Non-Identity)

This LoRA is a manga artist–focused rendering style LoRA, trained to emphasize how things are drawn, not who is drawn.

To avoid identity locking and average-face artifacts, the training data is intentionally distributed across many characters and complex compositions.

Training overview

• Characters: 43 female / 28 male

• Images per character: ~5–10

• Additional ~100 images with low identity emphasis
  (dynamic poses, complex layouts, manga panels, expressive linework)

Tagging

• Identity-focused tags were intentionally removed

• Captions prioritize line quality, rendering state, and composition

Usage

• Recommended weight: 0.8 – 1.2

  • Lower values → earlier / rougher style

  • Higher values → later / cleaner style

⚠️ This LoRA is not intended for reproducing specific characters.
Best used for style-driven manga rendering with flexible identity.