📝 简介内容 (Description)

Welcome to the post-apocalyptic potato world! This LoRA faithfully replicates the unique, melancholic, yet undeniably adorable art style of Tsukumizu's Girls' Last Tour.

🔑 Trigger Words

• Global Style Trigger: glt_style (Always include this to activate the post-apocalyptic sketch and coloring style).

• Character Summons: chito or yuuri . (Tip: Add their respective hair color tags for even more stable generations).

✨ Incredible Versatility & Crossover Potential This model isn't just for drawing Chito and Yuuri; its style generalization is exceptionally powerful! You can seamlessly pull characters from entirely different universes—like Genshin Impact's Raiden Shogun or Yae Miko—into this melancholic wasteland. The model applies a perfect "chibi, round-faced, flat-colored" filter. It makes even the most imposing characters look remarkably cute and bewildered. It is absolutely perfect for cross-universe scenarios, such as generating an illustration of an almighty Shogun staring in deadpan shock as a bizarre "system panel" suddenly pops up in front of her.

💡 Usage Guide & Tips

• How to "Potato-fy" Other Characters (e.g., Raiden Shogun): Use the base trigger glt_style + character tags (e.g., raiden shogun, long purple hair). Crucial Step: You must add round face, wide-set eyes, flat color, thick lineart to your positive prompts. In your negative prompts, you must heavily suppress the base model's default anatomy by adding: mature, realistic, sharp eyes, detailed face, pointy chin.

• Adjusting Mouths and Expressions: The model was trained extensively on native ASCII emoticons to capture the soul of the original manga. To change mouth shapes and expressions, simply use these tags in your prompt: :o (for a surprised/slightly open mouth), :d (for a happy, silly open mouth), >o< (for complaining/yelling), or = = (for the classic deadpan stare).

🛠️ Training Craftsmanship This model is the result of a highly rigorous training process. The dataset consists of 1349 meticulously curated high-definition screenshots, strictly divided into 4 distinct categories (Solo, Multiple, Scenery, and Close-ups) to prevent concept bleeding. The tagging process utilized a hybrid architecture: WD14 was used for precise booru-style tags, while Qwen was deployed to generate rich, natural language captions. This allows the model to deeply understand complex lighting, spatial depth, and atmospheric moods, responding beautifully to natural language prompts.

欢迎来到充满绝望却又无比治愈的废土土豆世界！本 LoRA 完美还原了漫画家水水（Tsukumizu）在《少女的终末旅行》中那种极具辨识度的粗粝线条与忧郁、呆萌并存的独特美术风格。

🔑 触发词 (Trigger Words)

• 全局画风触发词： glt_style（请始终将其放在提示词前排，以激活废土手绘线条与冷灰色调）。

• 专属角色召唤： chito 或 yuuri 。（提示：配合她们各自的发色标签，召唤效果更稳定）。

✨ 极强的泛用性与“跨宇宙”潜力 这个模型绝不仅仅只能用来画千户和尤莉，它的“画风滤镜”泛用性达到了惊人的地步！你可以极其丝滑地将其他宇宙的角色（例如《原神》的雷电将军或八重神子）直接拉进这个废土世界。模型会对她们进行强烈的“降维萌化打击”，赋予她们标志性的圆脸、宽眼距和平涂色彩。无论是用来做世界观碰撞的同人插图，还是用来描绘威严的雷神突然看到眼前弹出“系统面板”时的错愕与呆萌，都具有极强的喜剧效果与张力。

💡 核心使用指南

• 如何“终末旅行化”其他角色（以雷电将军为例）： 使用基础画风词 glt_style + 角色特征词（如 raiden shogun, long purple hair）。 最关键的一步： 你必须在正向提示词中加入 round face, wide-set eyes, flat color, thick lineart，同时在反向提示词中死死压制底模的写实权重，填入：mature, realistic, sharp eyes, detailed face, pointy chin。

• 嘴部与表情的微调魔法： 模型在训练时专门保留并强化了原著精髓的颜文字特征。想要调整人物的嘴型或表情，只需直接输入这些颜文字标签：:o（微张嘴/错愕），:d（开心/傻乎乎地张嘴），>o<（大声吐槽/气鼓鼓），或者 = =（经典的死鱼眼无语表情）。

🛠️ 炼丹工艺与技术细节 本模型采用了极度严苛的数据清洗与打标工艺。训练集包含 1349 张精挑细选的高清剧照，并在物理层面上严格划分为 4 大分类（单人特写、多人互动、纯大场景、物品特写）以杜绝概念污染。打标阶段采用了“双轨制”：使用 WD14 提取精确的二次元特征词，同时接入千问大模型进行深度的自然语言（Natural Language）打标。这使得模型对画面景深、光影走向以及废土氛围有着极深的理解，当你使用长句自然语言进行生图时，效果会异常惊艳。