From 3abc2514c9f9545debd3476a89907b375337c181 Mon Sep 17 00:00:00 2001
From: dseditor <dseditor@gmail.com>
Date: Sun, 4 Jan 2026 21:09:52 +0800
Subject: [PATCH] =?UTF-8?q?=E6=BA=96=E5=82=99=20LLM=20=E7=AF=80=E9=BB=9E?=
 =?UTF-8?q?=E6=95=B4=E5=90=88=EF=BC=9A=E5=89=B5=E5=BB=BA=E5=85=B1=E7=94=A8?=
 =?UTF-8?q?=20JSON=20=E6=8F=90=E5=8F=96=E6=A8=A1=E7=B5=84?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增 json_prompt_extractor.py 共用模組
- 提供 extract_prompts_from_json 函數
- 提供 should_extract_prompts 判斷函數
- 新增 LLM_JSON_INTEGRATION.md 整合說明
- 準備整合到所有 LLM 節點
---
 json_prompt_extractor.py | 101 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 101 insertions(+)
 create mode 100644 json_prompt_extractor.py

diff --git a/json_prompt_extractor.py b/json_prompt_extractor.py
new file mode 100644
index 0000000..649decb
--- /dev/null
+++ b/json_prompt_extractor.py
@@ -0,0 +1,101 @@
+"""
+共用的 JSON 提示詞提取工具
+用於從 LLM 輸出的 JSON 中提取 image_prompt 列表
+"""
+
+import json
+import re
+
+
+def extract_prompts_from_json(json_text):
+    """
+    從 JSON 文本中提取所有 image_prompt
+    
+    Args:
+        json_text: LLM 輸出的 JSON 文本
+        
+    Returns:
+        list: 提示詞列表，如果提取失敗返回空列表
+    """
+    try:
+        if not json_text or not json_text.strip():
+            return []
+        
+        # 清理 markdown 代碼塊標記
+        cleaned_text = json_text.strip()
+        if cleaned_text.startswith("```json"):
+            cleaned_text = cleaned_text[7:]
+        elif cleaned_text.startswith("```"):
+            cleaned_text = cleaned_text[3:]
+        
+        if cleaned_text.endswith("```"):
+            cleaned_text = cleaned_text[:-3]
+        
+        cleaned_text = cleaned_text.strip()
+        
+        # 嘗試解析 JSON
+        try:
+            data = json.loads(cleaned_text)
+        except json.JSONDecodeError:
+            # 如果解析失敗，嘗試提取 JSON 部分
+            start_idx = cleaned_text.find('{')
+            end_idx = cleaned_text.rfind('}') + 1
+            if start_idx != -1 and end_idx != 0:
+                json_str = cleaned_text[start_idx:end_idx]
+                data = json.loads(json_str)
+            else:
+                print("⚠️ 無法從輸出中提取 JSON")
+                return []
+        
+        # 提取所有 image_prompt
+        prompts = []
+        
+        # 1. 封面提示詞
+        if isinstance(data, dict) and "cover" in data:
+            cover = data["cover"]
+            if isinstance(cover, dict) and "image_prompt" in cover:
+                prompt = cover["image_prompt"]
+                if isinstance(prompt, str) and prompt.strip():
+                    prompts.append(prompt.strip())
+        
+        # 2. 內頁提示詞
+        if isinstance(data, dict) and "pages" in data:
+            pages = data["pages"]
+            if isinstance(pages, list):
+                for page in pages:
+                    if isinstance(page, dict) and "image_prompt" in page:
+                        prompt = page["image_prompt"]
+                        if isinstance(prompt, str) and prompt.strip():
+                            prompts.append(prompt.strip())
+        
+        # 3. 故事頁提示詞
+        if isinstance(data, dict) and "story_page" in data:
+            story = data["story_page"]
+            if isinstance(story, dict) and "image_prompt" in story:
+                prompt = story["image_prompt"]
+                if isinstance(prompt, str) and prompt.strip():
+                    prompts.append(prompt.strip())
+        
+        return prompts
+        
+    except Exception as e:
+        print(f"⚠️ 提取提示詞時發生錯誤: {e}")
+        return []
+
+
+def should_extract_prompts(output_text):
+    """
+    判斷輸出是否包含 JSON 格式的雜誌數據
+    
+    Args:
+        output_text: LLM 輸出文本
+        
+    Returns:
+        bool: 如果看起來像雜誌 JSON 則返回 True
+    """
+    if not output_text:
+        return False
+    
+    # 檢查是否包含關鍵字段
+    keywords = ["magazine_info", "cover", "pages", "image_prompt"]
+    return any(keyword in output_text for keyword in keywords)