这是我第一次尝试训练的二次元眼部微调模型。该 LoRA 专注于提升动漫角色的眼部光影层次、通透质感与微观细节，赋予眼睛如玻璃珠般晶莹剔透的水润光泽、细腻的放射状虹膜纹理与星芒高光。
⚠️ 温馨提示：由于是初次尝试训练，直出效果可能不够稳定，强烈建议配合局部重绘（Inpainting / ComfyUI Detailer）使用以获得最佳质感！
✨ 它擅长什么？/ What it is good at
晶莹通透的虹膜质感：呈现多层次的灰蓝/冷色调渐变，告别扁平单调的死色块。
微观反光与高光：能够生成细腻的十字星芒、水润高光点（Catchlights）与放射状光纹。
清晰的眼眶与睫毛：强化眼线轮廓与根根分明的睫毛边缘。
特写微距表现：在眼部微距特写（Extreme Close-up）下拥有惊艳的细节密度。
🎯 推荐使用场景 / What it should be used for
面部/眼部局部重绘（FaceDetailer / EyeDetailer / Inpainting）：用于修复底模生成中发糊、发白或缺乏生气的眼睛。
角色面部超近距离特写（Close-up Portraits）。
二次元插画的最终精修与画质提升。
⚠️ 它不擅长什么？/ What it is bad at
中远景/全身文生图（Text2Img）：当人物在画面中占比较小、脸部像素不足时，模型可能无法展现微观线条，甚至导致视线轻微偏移。
写实真人风格（Photorealistic）：本模型专为二次元动漫（Anime/2D）美学设计，不适用于真实摄影质感。
极度夸张的异形眼眶：对非标准二次元眼眶结构的泛化性较弱。
❌ 不建议的使用方式 / How it should NOT be used
不建议直接用于全身/大中景文生图直接出片（此时眼部像素过小，下采样会抹平细节）。
不建议在局部重绘时开启过高的重绘幅度（Denoise > 0.85），以免双眼独立重绘时导致视线不对称或瞳孔轻微漂移。
Model Overview
This is my first experimental LoRA model, designed to enhance the lighting, translucency, and microscopic details of anime-style eyes. It adds a luminous, crystal-clear glass-like texture, intricate radiating iris patterns, sharp eyelashes, and sparkling catchlights to character portraits.
⚠️ Note: Since this is a v1.0 experimental release, direct text-to-image results might be inconsistent. It is strongly recommended to use this LoRA via Inpainting / ComfyUI FaceDetailer workflows for optimal results!
✨ What is this model good at?
Translucent & Radiant Irises: Adds deep, multi-tonal gray-blue gradients and crystal reflections instead of flat color blocks.
Micro Catchlights & Radiating Lines: Renders delicate starburst highlights and glowing optical details inside the pupil.
Sharp Lash & Eyeliner Definition: Accentuates clean, stylized anime eyelashes and contours.
Macro & Close-up Focus: Delivers breathtaking clarity in high-resolution eye close-ups.
🎯 What should it be used for?
Local Inpainting / Eye & Face Detailer: Perfect as a refiner pass to upgrade blurry or dull eyes generated by base models.
Extreme Close-up Portraits & Character Sheets.
Post-processing and quality refinement for anime illustrations.
⚠️ What is it bad at? / Limitations
Full-body / Wide-angle Text2Image: When the character’s face is small on the canvas, micro-details will be lost during downscaling, and pupils may drift.
Photorealistic / Real-life Eyes: Specifically tailored for 2D/Anime aesthetics; incompatible with realistic photography.
Extreme Non-standard Eye Shapes: Generalization on highly stylized monster or non-human eyes is limited.
❌ How should it NOT be used?
Do NOT rely on it for small-face full-body generation from scratch (the lack of canvas pixels will smudge fine iris lines).
Do NOT use overly high Inpaint Denoise (> 0.85) in dual-eye workflows, as independent sampling may lead to slight pupil misalignment or strabismus.