QY_Z_Anime_Style Lora

QY_STUDIO
8
47
stylensfwloracompatiblez imagez animeaesthetic animez anime lora
v1.0
model-image-0
model-image-1
model-image-2
model-image-3
model-image-4
model-image-5
model-image-6
model-image-7
model-image-8
model-image-9
model-image-10

详情

下载文件 (1)

模型描述

QY_Z_Anime_Style Lora 禁止搬运至哩布哩布这种商业类型平台！！！！！！！！

【中文介绍】

模型名称：QY_Z_Anime_Style Lora

模型类型：专属原创动漫画风LoRA

适配底模：Z_anime bf16 base（Z_image 变体模型）SeeSee21/Z-Anime · Hugging Face

训练步数：6000步 精细训练

推荐权重：0.9~1.2（可突破1.0上限，最高1.2稳定使用，无结构崩坏）

✨ 模型核心亮点

本LoRA是一款纯画风改造型模型，核心主打零干预结构、只改画风的优质特性，规避了大多数画风LoRA篡改人物动作、画面构图、场景结构的问题。可完整保留原有提示词生成的人物姿态、镜头角度、场景布局，仅统一替换画面画风，即使权重拉至1.2也不会对画面内容、整体构图造成破坏性影响。同时全面支持NSFW生成场景，适配性更广，二次元常规及私密创作场景均可稳定适配。

模型经过6000步足量精细迭代训练，训练数据集均为手工精修打磨。针对AI绘画常见缺陷做了专项优化，逐图补全手脚细节、矫正肢体结构、修复画面瑕疵，大幅降低手脚崩坏、结构错乱、细节缺失的概率，出图稳定性极高、画面规整精致。

模型拥有干净细腻的动漫手绘质感，线条流畅、色调柔和通透，画面氛围感统一自然，不会出现风格杂乱、画面割裂的情况，出图观感干净高级。

📌 使用小贴士

  • 权重适配性极强，支持0.9-1.2权重区间，可突破1.0上限，高权重下风格浓郁自然，不破坏画面构图、不产生畸变与内容崩坏

  • 仅适配 Z_anime 系列底模，最优适配 Z-Anime_distill8Step 蒸馏版本，同时兼容所有以 Z_anime 为基底的蒸馏模型，搭配对应底模出图效果最佳、风格还原度百分百

  • 通用二次元场景、人物立绘、插画、氛围感壁纸等场景均可适配，万能百搭不挑题材

  • 提示词使用说明：支持中文、英文自然语言自由描述，无固定触发词，可直接通过LoRA加载器启用，仅生效画风迁移，无需额外适配关键词

⚙️ 精准生图参数推荐

  • Z_anime bf16 base 原版模型：采样次数 28-45、采样器 euler_ancestral、调度器 beta、CFG 推荐 4.5

  • Z-Anime_distill8Step 蒸馏模型：采样次数 8、采样器 euler_ancestral、调度器 beta、CFG 推荐 1.0-1.2

【English Introduction】

Model Name：QY_Z_Anime_Style Lora

Model Type：Original Exclusive Anime Style LoRA

Base Model：Z_anime bf16 base (Variant of Z_image) SeeSee21/Z-Anime · Hugging Face

Training Steps：6000 Steps Fine Training

Recommended Weight：0.9~1.2 (Stable beyond 1.0, max 1.2 without structural damage)

✨ Key Features

This is a pure style-customized LoRA for anime creation. Its core advantage is only changes the painting style without affecting character poses, compositions and scene structures. It completely retains the original prompt’s design of characters, camera angles and scene layout, and unifies the overall image texture. Even at 1.2 weight, it will not cause major damage to image composition and content. Fully supports NSFW generation, suitable for various anime creation scenarios.

Trained with 6000 sufficient iterative steps, the entire training dataset is manually refined. We specially optimized AI common generation problems, completed hand and foot details, corrected limb structure errors and fixed image defects one by one, which greatly reduces limb distortion and structural collapse, ensuring stable and exquisite output results.

It delivers clean and delicate hand-painted anime texture with smooth lines and soft tones. The overall picture atmosphere is unified and natural, without messy style mixing or image fragmentation, bringing high-quality and consistent visual effects.

📌 Usage Guide

  • Excellent weight compatibility,stable at 0.9-1.2 weight range, supports over 1.0 with no image distortion or structural damage

  • Only compatible with Z_anime series base models. Best performance with Z-Anime_distill8Step distilled version, also compatible with all Z_anime-based distilled models for perfect style restoration

  • Widely compatible with all anime scenarios: character portraits, illustrations, atmosphere wallpapers, original artworks and more

  • Prompt Guide: Supports both Chinese and English natural language prompts. No trigger words required. Simply load the LoRA via the model loader for direct use.

⚙️ Recommended Generation Parameters

  • Z_anime bf16 base: Sampling Steps 28-45, Sampler euler_ancestral, Scheduler beta, CFG Scale 4.5

  • Z-Anime_distill8Step: Sampling Steps 8, Sampler euler_ancestral, Scheduler beta, CFG Scale 1.0-1.2

此模型生成的图像