图像工作台

模型

提示词2,970 / 1,000

{
  "type": "可爱教育科普信息图海报",
  "topic": "[大语言模型的训练过程]",
  "subtitle": "[从海量数据中学习，变成“超会聊天”的小助手！]",
  "style": {
    "overall": "萌系中文科普解说海报，马卡龙低饱和课堂信息图，圆角卡片分栏，柔和米色背景，手绘卡通美学，利落矢量插画风格，温馨亲切的科普语调",
    "palette": ["米白", "薰衣草紫", "天青蓝", "薄荷绿", "暖橙黄", "淡粉", "柔和棕"],
    "rendering": "高品质扁平插画配微弱柔和阴影，清晰可辨的中文排版，贴纸质感点缀，细小星芒、爱心、箭头与对话气泡"
  },
  "layout": {
    "format": "横版海报",
    "grid": {"rows": 2, "columns": 4, "count": 8},
    "sections": [
      {"title": "1. 数据收集", "position": "top-left", "labels": ["网页", "新闻", "对话"]},
      {"title": "2. 数据预处理", "position": "top row, second column", "labels": ["今天 天气 真不错!!", "今天|天气|真|不错!!"]},
      {"title": "3. 预训练", "position": "top row, third column", "labels": ["今天气", "很好", "不错", "?", "......"]},
      {"title": "4. 监督微调（SFT）", "position": "top-right", "labels": ["问：太阳为什么会发光？", "答：因为……", "好回答！"]},
      {"title": "5. 奖励模型训练（RM）", "position": "bottom-left", "labels": ["回答A 更好!", "回答B 一般般"]},
      {"title": "6. 强化学习（RLHF）", "position": "bottom row, second column", "labels": ["奖励 +1", "惩罚 -1"]},
      {"title": "7. 评估与测试", "position": "bottom row, third column", "labels": ["知识能力", "推理能力", "安全性", "稳定性"]},
      {"title": "8. 部署与应用", "position": "bottom-right", "labels": ["聊天", "写代码", "写文章"]}
    ],
    "topDecorations": ["左上角微光闪烁", "主标题周围粉色碎花涂鸦线", "右上角白色小吉祥物手持星星魔法棒", "对话框写着“冲鸭!”"],
    "bottomDecorations": ["左下角系蝴蝶结的白色小吉祥物", "底部7个圆角小方框带箭头的流程总结长条", "居中底部总结句", "右下角写有鼓励文字带爱心的便签纸"]
  },
  "characters": {
    "main_mascots": ["粉红脸颊的圆润白熊吉祥物", "深色面屏搭配绿色天线的小圆机器人", "扎马尾的人类女老师", "手拿放大镜的毛茸茸白色小考官"],
    "recurring_robot_design": "矮胖可爱机器人，圆润机身，浅绿与米白外壳，深蓝发光眼睛屏幕，短小四肢"
  },
  "sectionDetails": [
    {"title": "1. 数据收集", "panelColor": "lavender", "scene": "戴紫色帽子的小白熊坐在成堆彩色书籍旁使用笔记本电脑，周围漂浮圆角标签代表网络数据源", "caption": "数据越多，知识越丰富！"},
    {"title": "2. 数据预处理", "panelColor": "sky blue", "scene": "戴蓝色帽子的小吉祥物正把零碎纸屑扫进桶里，箭头将凌乱未清洗的句子转化成规范分词后的干净文本", "caption": "把脏乱差，变得整整齐齐～"},
    {"title": "3. 预训练", "panelColor": "mint green", "scene": "小机器人正在阅读一本翻开的绿色大书，周围气泡展示基础字词分词与预测，生动表现语言自主学习", "caption": "像小朋友学习说话一样！"},
    {"title": "4. 监督微调（SFT）", "panelColor": "golden yellow", "scene": "老师指着带标签的问答卡片授课，吉祥物认真听讲，突出高质量人标问答对", "caption": "老师带着学，回答更靠谱！"},
    {"title": "5. 奖励模型训练（RM）", "panelColor": "pink", "scene": "小机器人站在绿色对勾和红色叉号之间，对比两个回答选项，学会评判优劣", "caption": "学会“挑好答案”！"},
    {"title": "6. 强化学习（RLHF）", "panelColor": "lavender-blue", "scene": "机器人手拿工具，积分板通过箭头展示得分奖惩反馈", "caption": "鼓励好的，改掉不好的！"},
    {"title": "7. 评估与测试", "panelColor": "blue", "scene": "毛茸茸考官手拿放大镜，旁边记录板打有四个绿色通过标记", "caption": "全方位体检，确保质量过关！"},
    {"title": "8. 部署与应用", "panelColor": "soft red-pink", "scene": "机器人被聊天、写作、编程、文档等应用图标环绕，呈现真实落地场景", "caption": "正式上岗，陪你聊天写作啦！"}
  ],
  "bottomSummary": {
    "title": "总结一下：",
    "steps": ["数据收集 打好基础", "预处理 整理数据", "预训练 学知识", "监督微调 学会回答", "奖励模型 学会评判", "强化学习 对齐人类偏好", "评估测试 质量把关"],
    "closingText": "[这就是大语言模型从“啥也不懂”到“超级学霸”的成长之路！٩(๑^o^๑)۶]"
  },
  "language": "Simplified Chinese",
  "quality": "适合社交媒体传播的高清科普图，间距均衡，排版清晰美观，生动吸睛"
}

Ctrl / ⌘ + Enter

参考图0 / 16

画幅

画质

分辨率

图像数量

1 / 4

结果

从一句提示词开始

生成结果会展示在这里。

获取灵感