Yoneyama Mai | 米山舞

Ethereal anime aesthetic for Illustrious-XL | Trigger: yoneyama_mai

✨ Overview

This LoRA is fine-tuned on Illustrious-XL v0.1 to replicate the signature ethereal digital anime style inspired by artist Yoneyama Mai. It excels at generating dreamy, soft-toned illustrations with delicate linework, atmospheric lighting, and pastel color harmony.

🎭 Style Highlights

• 🌸 Pastel Macaron Palette: Low-saturation tones with smooth gradient transitions

• 💫 Ethereal Glow: Soft lighting, light diffusion, and dreamy ambient effects

• 👁️ Delicate Rendering: Refined linework, expressive eyes, and subtle character details

• 🌌 Abstract Composition: Floating particles, bokeh backgrounds, subject-focused framing

🔑 Trigger Word

• Primary: yoneyama_mai

→ Place at the beginning of your prompt, recommended weight: 1.0

🎨 Prompt Template

yoneyama_mai, digital art, anime style, soft lighting, ethereal, pastel colors, [your character/scene description]

⚙️ Recommended Settings

| Parameter | Value | Note |

|-----------|-------|------|

| Base Model | Illustrious-XL v0.1 | Required — not tested on other XL variants |

| Resolution | 1024x1024 | SDXL native resolution; other official sizes supported |

| Sampler | DPM++ 2M Karras | Best for detail fidelity & sharp linework |

| Sampler | Euler a | Preferred for softer, more aesthetic/painterly results |

| CFG Scale | 6.5 ~ 7.5 | >8 may overfit; <6 weakens style influence |

| Steps | 25 ~ 30 | Balanced quality & inference speed |

| LoRA Weight | 0.8 ~ 1.0 | Full strength recommended for pure style transfer |

💡 Pro Tips

• 🔍 More Details: Use DPM++ 2M Karras + CFG 7.0 for crisp linework and fine features

• 🎨 More Aesthetic: Try Euler a + slightly lower CFG (6.5) for dreamy, artistic vibes

• 🧩 ControlNet Ready: Compatible with lineart/depth ControlNets for pose/structure control (no specific preset required)

• 🌐 English Only: Optimized for English prompts; other languages may reduce style consistency

• 📐 Resolution Matters: Stick to 1024x1024 or SDXL-native aspect ratios to avoid distortion

⚠️ Limitations & Usage Notes

• Trained on 20 Danbooru-sourced images with auto-generated tags (JoyCaption); manual curation minimal

• Hands, text, and complex structures may require post-processing or ControlNet assistance

• Avoid mixing with conflicting keywords: photorealistic, 3d render, oil painting, ugly, bad anatomy

• Best for single-character illustrations; multi-character or complex scenes may have limited generalization

• 🔒 Personal Research Use Only — Commercial use, artist impersonation, or sensitive content generation is strictly prohibited

🔄 Future Plans

Potential updates may include:

• Compatibility with other Illustrious-XL variants

• Pony diffusion model support

• Expanded dataset for improved multi-scene generalization

Model fine-tuned with LoRA (dim=16, alpha=8) on Illustrious-XL v0.1 | fp16 precision | English prompts optimized