RUNA V2
Details
Download Files
About this version
Model description
Girl combination of Asian-Japanese-Indonesian-Arabic-Korean
Perfect 👍 I’ll draft a professional English description for your RUNA V2 LoRA so it’s ready for CivitAI review. This will make it more appealing, clear, and aligned with how CivitAI expects model documentation.
📌 Model Description
Name: RUNA V2
Type: LoRA
Base Model: Flux.1 D
RUNA V2 is a character-focused LoRA designed to create unique female portraits blending features from multiple cultural backgrounds: Asian, Japanese, Indonesian, Arabic, and Korean. This combination results in characters with distinctive and versatile appearances—balancing realism with stylized beauty.
It is suitable for generating:
Realistic or semi-realistic female characters
Multicultural character designs for stories or concept art
Portraits with elegant, soft, and expressive features
Hybrid looks that combine different Asian and Middle Eastern aesthetics
🛠️ Training Details
Steps: 20
Epochs: 4
Clip Skip: 1
Recommended Strength: 0.8
AutoV2 Hash: DF36D36996
Trained with a curated dataset of portraits across the represented cultural backgrounds, ensuring diverse facial structures, skin tones, and styles while maintaining consistent anatomy and natural beauty.
✨ Key Features
Hybrid appearance combining Asian & Middle Eastern traits
Natural facial proportions with expressive realism
Works well in both realistic and anime-inspired outputs
Good versatility for portrait, fashion, and storytelling use
🎨 Usage Guide
Recommended with Flux.1 D as the base model.
Apply LoRA at 0.7–0.9 strength for balanced results.
Prompting suggestions:
Positive Prompt: beautiful woman, multicultural features, elegant portrait, soft lighting, natural look
Optional Enhancements: high detail, cinematic style, smooth skin, 8k portrait, subtle makeup
Negative Prompt: deformed, blurry, extra limbs, low quality, unnatural face
Combine with ControlNet or style LoRAs for creative flexibility.
⚠️ Notes
Made for artistic and creative projects only.
Avoid misuse such as deepfakes or harmful applications.
Output quality depends on prompts, sampling methods, and the chosen base model.
👉 Do you also want me to rebuild optimized tag sets (like “Realistic Portrait, Multicultural, Female Character, Semi-Realistic, Photographic”) so both this and your previous LoRA get better search visibility on CivitAI?

