Realistic Vision Style LoRA for Z-Image

Model Details

• Base Model: Z-Image DE-Turbo v1
• Training: 20 epochs, 100 images
• Network: LoRA (dim=64, alpha=32)
• Original Style: Realistic Vision from CivitAI

Trigger Word

Usage

• 1girl, portrait, long black hair, soft lighting, realistic_vision_style
• 1boy, casual outfit, modern aesthetic, realistic_vision_style

Sample Images

Training Info

• Dataset: 100 images generated with original CivitAI model
• VLM Captioning: Qwen2-VL-7B
• Learning Rate: 5e-5
• Optimizer: AdamW8bit