樱岛麻衣Sakurajima Mai
详情
下载文件
模型描述
模型简介：
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。
Model Introduction：
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.
基础参数:
- 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。
Basic parameters:
- 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.
核心参数:
- 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。
Core parameters:
- 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.
效果说明:
- 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 ）。
Effect description:
- 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).
提示词示例（可根据需求增减细节）:
Prompt word example (details can be added or reduced as needed):
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 ）
注意事项:
- 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。
Notes:
- Extreme parameters (such as CFG>12, LoRA weight>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.