<h3 id="-xxx1n6x63">模型简介：</h3><h3 id="lora-9brrc4i0q">基于《青春猪头少年不会梦到兔女郎学姐》中 樱岛麻衣 角色，训练的专属 LoRA 模型。聚焦还原她的经典形象（黑长直发型、校服/兔女郎等服饰特征），适配动漫风格底模，轻松生成贴合原作气质的画面。</h3><h3 id="model-introduction-qv3efsrbq">Model Introduction：</h3><h3 id="based-on-the-character-of-mai-sakurajima-in-&quot;rascal-does-not-dream-of-bunny-girl-senpai&quot;-we-trained-a-dedicated-lora-model.-we-focused-on-restoring-her-classic-image-(long-straight-black-hair-school-uniformbunny-girl-and-other-clothing-features)-adapted-the-anime-style-base-model-and-easily-generated-a-picture-that-fits-the-original-temperament.-xonte9zcf">Based on the character of Mai Sakurajima in "Rascal Does Not Dream of Bunny Girl Senpai", we trained a dedicated LoRA model. We focused on restoring her classic image (long straight black hair, school uniform/bunny girl and other clothing features), adapted the anime style base model, and easily generated a picture that fits the original temperament.</h3><h3 id=":-eo6pb4p23">基础参数:</h3> - 底模选择：推荐搭配 AnimeSharp/AnythingV5 等动漫向 Checkpoint，风格适配度更高。- 采样器：优先尝试  Euler a  /  DPM++ 2M Karras ，画面细腻度与风格还原更平衡。<h3 id="basic-parameters:-zib4mxqb0">Basic parameters:</h3>- Base model selection: It is recommended to use AnimeSharp/AnythingV5 and other animation-oriented Checkpoints, which are more suitable for the style.- Sampler: Try Euler a / DPM++ 2M Karras first, which has a better balance between picture detail and style restoration.<h3 id=":-ice1q8kn3">核心参数:</h3> - LoRA 权重：建议  0.6 - 0.8 （强度过高易导致面部崩坏，过低则风格不明显，可根据提示词灵活微调）。- CFG Scale：推荐  6 - 8 （平衡提示词约束与模型创意，过高可能让画面僵硬，过低会偏离角色特征）。- Denoising Strength（重绘强度，用于 img2img）：若基于原图二次创作， 0.4 - 0.6  可保留原图基础，同时融入模型风格；纯文生图无需此参数。<h3 id="core-parameters:-z25ou6cnz">Core parameters:</h3>- LoRA weight: recommended 0.6 - 0.8 (too high strength can easily lead to facial collapse, too low strength will make the style unclear, can be flexibly adjusted according to the prompt word).- CFG Scale: recommended 6 - 8 (balance prompt word constraints and model creativity, too high may make the picture stiff, too low will deviate from the character characteristics).- Denoising Strength (redrawing strength, used for img2img): If based on the original image for secondary creation, 0.4 - 0.6 can retain the original image foundation while incorporating the model style; pure text images do not need this parameter.<h3 id=":-d9397eicd">效果说明:</h3> - 文生图：输入角色+服饰关键词（如  school uniform  /  bunny girl suit  ），可生成符合原作风格的新场景，发型、神态还原度较高。- 图生图：对动漫截图、同人草稿重绘，能强化“麻衣学姐”风格，优化画面细节（如校服纹理、发丝质感 ）。<h3 id="effect-description:-2mysk220q">Effect description:</h3>- Text-based pictures: Input character + clothing keywords (such as school uniform / bunny girl suit) to generate new scenes that match the style of the original work, with a high degree of restoration of hairstyles and expressions.- Picture-based pictures: Redraw anime screenshots and fan drafts to enhance the style of "Mai Senior Sister" and optimize picture details (such as school uniform texture and hair texture).<h3 id=":-3exqkeqrt">提示词示例（可根据需求增减细节）:</h3><h3 id="prompt-word-example-(details-can-be-added-or-reduced-as-needed):-kupz9wnki">Prompt word example (details can be added or reduced as needed):</h3>一、核心角色特征(Core Character Characteristics) - 基础形象(Basic image)：mayi, Sakurajima Mai, long black hair, school uniform, pink bow （麻衣，樱岛麻衣，黑色长发，校服，粉色领结 ）- 经典元素(Classic elements)： bunny girl suit, black stockings, rabbit hairpin （兔女郎套装，黑丝，兔耳发卡 ）- 神态表情(Expression)： calm expression, slightly smile, lazy charm （冷静神情，微微笑容，慵懒气质 ） 二、画面风格/氛围(Graphic style/atmosphere) - 原作还原(Original restoration)： anime style, Seishun Buta Yarou, official art （动漫风格，青春猪头少年，官方画风 ）- 艺术强化（Art Enhancement）： watercolor, ink painting, oil painting （水彩风，水墨风，油画质感 ）- 场景氛围(Scene atmosphere)： school classroom, sunset glow, city night （学校教室，夕阳余晖，城市夜景 ） 三、细节补充(Additional details) - 画质提升(Improved image quality)： best quality, ultra-detailed, 8k resolution （最佳质量，极致细节，8K 分辨率 ）- 动态动作（Dynamic Action)： sitting pose, walking, holding rabbit plush （坐姿，漫步，抱兔玩偶 ）- 特殊效果(Special Effects)： glowing eyes, dreamy atmosphere, lens flare （眼神发光，梦幻氛围，镜头光晕 ） 四、负面提示词(Negative prompt words) lowres, bad anatomy, deformed, extra limbs, text, watermark （低分辨率，畸形人体，多余肢体，文字水印 ）<h3 id=":-ud96ppazh">注意事项:</h3> - 极端参数（如 CFG＞12、LoRA 权重＞1.0 ）可能引发画面畸变，建议循序渐进调整。- 搭配更详细的提示词（如  detailed face   cinematic lighting  ），可进一步提升画面精致度。<h3 id="notes:-nwpg3ok1a">Notes:</h3>- Extreme parameters (such as CFG&gt;12, LoRA weight&gt;1.0) may cause image distortion, so it is recommended to adjust gradually.- With more detailed prompts (such as detailed face cinematic lighting), the image refinement can be further improved. <h2 id="&quot;&quot;-h8vskggq9">欢迎大家测试反馈，一起探索更多“麻衣学姐”的创作玩法～</h2><h2 id="welcome-everyone-to-test-and-give-feedback-and-explore-more-creative-ways-to-play-&quot;mai-senior-sister&quot;-together~-cq0gzz118">Welcome everyone to test and give feedback, and explore more creative ways to play "Mai Senior Sister" together~</h2>

Files with SHA256: 15d80d539a5910e837ed69d51aa65518ccf4fbc2d467ff2529636de93b826845