# GPT Image Hub prompt

## 可愛大型語言模型訓練科普圖解
- URL: https://www.gptimagehub.com/zh-TW/library/cute-llm-training-educational-infographic
- Generator URL: https://www.gptimagehub.com/zh-TW/generate?promptId=oss-ym-g2-14665
- Markdown URL: https://www.gptimagehub.com/api/agent/markdown?path=%2Fzh-TW%2Flibrary%2Fcute-llm-training-educational-infographic
- Category: 資訊圖與教育
- Model: gpt-image-2
- Aspect ratio: 16:9
- Style: illustration
- Tags: 資訊圖表, 大模型訓練, 科普海報, 可愛吉祥物, 粉彩插畫, 向量風格
- Updated: 2026-10-27T01:47:58.098Z
- Created: 2026-10-27T01:47:58.098Z

生成一張粉嫩馬卡龍色系的中文科普海報，以八格漫畫圖解大語言模型的完整訓練流程。可靈活調整標題、標籤文字與吉祥物以解說各類AI主題。

### Full prompt
```text
{
  "type": "可愛教育科普資訊圖海報",
  "topic": "[大語言模型的訓練過程]",
  "subtitle": "[從海量資料中學習，變成“超會聊天”的小助手！]",
  "style": {
    "overall": "萌系中文科普解說海報，馬卡龍低飽和課堂資訊圖，圓角卡片分欄，柔和米色背景，手繪卡通美學，利落向量插畫風格，溫馨親切的科普語調",
    "palette": ["米白", "薰衣草紫", "天青藍", "薄荷綠", "暖橙黃", "淡粉", "柔和棕"],
    "rendering": "高品質扁平插畫配微弱柔和陰影，清晰可辨的中文排版，貼紙質感點綴，細小星芒、愛心、箭頭與對話氣泡"
  },
  "layout": {
    "format": "橫版海報",
    "grid": {"rows": 2, "columns": 4, "count": 8},
    "sections": [
      {"title": "1. 資料收集", "position": "top-left", "labels": ["網頁", "新聞", "對話"]},
      {"title": "2. 資料預處理", "position": "top row, second column", "labels": ["今天 天氣 真不錯!!", "今天|天氣|真|不錯!!"]},
      {"title": "3. 預訓練", "position": "top row, third column", "labels": ["今天氣", "很好", "不錯", "?", "......"]},
      {"title": "4. 監督微調（SFT）", "position": "top-right", "labels": ["問：太陽為什麼會發光？", "答：因為……", "好回答！"]},
      {"title": "5. 獎勵模型訓練（RM）", "position": "bottom-left", "labels": ["回答A 更好!", "回答B 一般般"]},
      {"title": "6. 強化學習（RLHF）", "position": "bottom row, second column", "labels": ["獎勵 +1", "懲罰 -1"]},
      {"title": "7. 評估與測試", "position": "bottom row, third column", "labels": ["知識能力", "推理能力", "安全性", "穩定性"]},
      {"title": "8. 部署與應用", "position": "bottom-right", "labels": ["聊天", "寫程式碼", "寫文章"]}
    ],
    "topDecorations": ["左上角微光閃爍", "主標題周圍粉色碎花塗鴉線", "右上角白色小吉祥物手持星星魔法棒", "對話方塊寫著“衝鴨!”"],
    "bottomDecorations": ["左下角系蝴蝶結的白色小吉祥物", "底部7個圓角小方框帶箭頭的流程總結長條", "居中底部總結句", "右下角寫有鼓勵文字帶愛心的便籤紙"]
  },
  "characters": {
    "main_mascots": ["粉紅臉頰的圓潤白熊吉祥物", "深色面屏搭配綠色天線的小圓機器人", "扎馬尾的人類女老師", "手拿放大鏡的毛茸茸白色小考官"],
    "recurring_robot_design": "矮胖可愛機器人，圓潤機身，淺綠與米白外殼，深藍髮光眼睛螢幕，短小四肢"
  },
  "sectionDetails": [
    {"title": "1. 資料收集", "panelColor": "lavender", "scene": "戴紫色帽子的小白熊坐在成堆彩色書籍旁使用膝上型電腦，周圍漂浮圓角標籤代表網路資料來源", "caption": "資料越多，知識越豐富！"},
    {"title": "2. 資料預處理", "panelColor": "sky blue", "scene": "戴藍色帽子的小吉祥物正把零碎紙屑掃進桶裡，箭頭將凌亂未清洗的句子轉化成規範分詞後的乾淨文字", "caption": "把髒亂差，變得整整齊齊～"},
    {"title": "3. 預訓練", "panelColor": "mint green", "scene": "小機器人正在閱讀一本翻開的綠色大書，周圍氣泡展示基礎字詞分詞與預測，生動表現語言自主學習", "caption": "像小朋友學習說話一樣！"},
    {"title": "4. 監督微調（SFT）", "panelColor": "golden yellow", "scene": "老師指著帶標籤的問答卡片授課，吉祥物認真聽講，突出高質量人標問答對", "caption": "老師帶著學，回答更靠譜！"},
    {"title": "5. 獎勵模型訓練（RM）", "panelColor": "pink", "scene": "小機器人站在綠色對勾和紅色叉號之間，對比兩個回答選項，學會評判優劣", "caption": "學會“挑好答案”！"},
    {"title": "6. 強化學習（RLHF）", "panelColor": "lavender-blue", "scene": "機器人手拿工具，積分板透過箭頭展示得分獎懲反饋", "caption": "鼓勵好的，改掉不好的！"},
    {"title": "7. 評估與測試", "panelColor": "blue", "scene": "毛茸茸考官手拿放大鏡，旁邊記錄板打有四個綠色透過標記", "caption": "全方位體檢，確保質量過關！"},
    {"title": "8. 部署與應用", "panelColor": "soft red-pink", "scene": "機器人被聊天、寫作、程式設計、文件等應用圖示環繞，呈現真實落地場景", "caption": "正式上崗，陪你聊天寫作啦！"}
  ],
  "bottomSummary": {
    "title": "總結一下：",
    "steps": ["資料收集 打好基礎", "預處理 整理資料", "預訓練 學知識", "監督微調 學會回答", "獎勵模型 學會評判", "強化學習 對齊人類偏好", "評估測試 質量把關"],
    "closingText": "[這就是大語言模型從“啥也不懂”到“超級學霸”的成長之路！٩(๑^o^๑)۶]"
  },
  "language": "Simplified Chinese",
  "quality": "適合社交媒體傳播的高畫質科普圖，間距均衡，排版清晰美觀，生動吸睛"
}
```