""" 批量替換 photo_magazine_generator.py 中的中文訊息為英文 """ import re # 中文到英文的映射 translations = { # PhotoMagazinePromptGenerator "📸 檢測到參考圖片,建議使用 LLM 節點提取人物特徵": "📸 Reference image detected, recommend using LLM node to extract person features", "自動注入 {EXTRACT_FROM_IMAGE} 佔位符": "Auto-injecting {EXTRACT_FROM_IMAGE} placeholder", "模板:": "Template:", "模特兒:": "Model:", "風格:": "Style:", "場景:": "Scene:", "頁數:": "Pages:", "特徵:": "Features:", # PhotoMagazineParser "📝 開始解析 LLM 輸出的 JSON...": "📝 Starting to parse LLM JSON output...", "✅ 解析完成!提取到": "✅ Parsing complete! Extracted", "個圖片提示詞": "image prompts", "錯誤:JSON 輸入為空": "Error: JSON input is empty", "錯誤:": "Error:", "警告:未找到任何 image_prompt": "Warning: No image_prompt found", "解析錯誤:": "Parse error:", " ✓ 封面提示詞": " ✓ Cover prompt", " ✓ 頁面": " ✓ Page", "提示詞": "prompt", " ✓ 故事頁提示詞": " ✓ Story page prompt", # PhotoMagazineMaker "成功解析JSON,包含": "Successfully parsed JSON, contains", "個頁面": "pages", "開始轉換圖片,圖片類型:": "Starting image conversion, image type:", "轉換圖片": "Converting image", "轉換批次圖片": "Converting batch image", "轉換單張圖片:": "Converting single image:", "成功轉換": "Successfully converted", "張圖片": "images", "錯誤:沒有有效的圖片可以處理": "Error: No valid images to process", "總共需要": "Total required", "個頁面,可用圖片": "pages, available images", "圖片分配策略:": "Image allocation strategy:", "使用圖片": "Using image", "作為封面": "as cover", "警告:沒有圖片可用作封面": "Warning: No image available for cover", "警告:JSON中的cover不是字典格式(類型:": "Warning: cover in JSON is not dict format (type:", "),使用預設值": "), using default values", "警告:JSON中缺少cover數據,使用預設值": "Warning: Missing cover data in JSON, using default values", "✓ 關閉封面排版,使用第一張圖片作為滿版封面": "✓ Cover layout disabled, using first image as full bleed cover", "封面繪製完成": "Cover page complete", "JSON中有": "JSON contains", "個頁面數據,準備繪製所有內頁": "page data, preparing to draw all content pages", "作為第": "as page", "頁": "page", "頁使用備用圖片": "using fallback image", "警告:第": "Warning: Page", "頁數據格式錯誤,跳過": "data format error, skipping", "頁繪製完成": "page complete", "第": "Page", "警告:JSON中的story_page不是字典格式(類型:": "Warning: story_page in JSON is not dict format (type:", "警告:JSON中缺少story_page數據,使用預設值": "Warning: Missing story_page data in JSON, using default values", "作為故事頁": "as story page", "故事頁繪製完成": "Story page complete", "作為尾頁": "as back cover", "尾頁繪製完成": "Back cover complete", "PDF 生成成功!": "PDF generated successfully!", "檔案位置:": "File location:", "生成寫真雜誌時發生錯誤:": "Error generating photo magazine:", "JSON 解析錯誤:": "JSON parse error:", ",接收到的數據:": ", received data:", "JSON格式錯誤:期望字典格式,收到": "JSON format error: expected dict, received", "JSON格式錯誤:pages必須是列表格式,收到": "JSON format error: pages must be list, received", "JSON數據錯誤:pages列表為空,無法生成雜誌": "JSON data error: pages list is empty, cannot generate magazine", "圖片分配結果:": "Image allocation result:", "封面=": "cover=", "內頁=": "pages=", "故事頁=": "story=", "尾頁=": "footer=", "版型B顯示了": "Layout B displayed", "行文字,字體大小:": "lines of text, font size:", "行高:": "line height:", "框高:": "box height:", "最大行數:": "max lines:", "滿版圖片創建錯誤:": "Full bleed image creation error:", "圖片轉換錯誤:": "Image conversion error:", } def replace_chinese_messages(file_path): """替換文件中的中文訊息""" with open(file_path, 'r', encoding='utf-8') as f: content = f.read() # 按照從長到短的順序替換,避免部分匹配 sorted_translations = sorted(translations.items(), key=lambda x: len(x[0]), reverse=True) for chinese, english in sorted_translations: content = content.replace(chinese, english) with open(file_path, 'w', encoding='utf-8') as f: f.write(content) print(f"✓ Replaced Chinese messages in {file_path}") if __name__ == "__main__": file_path = r"f:\CUI\ComfyUI\custom_nodes\ComfyUI-ListHelper\photo_magazine_generator.py" replace_chinese_messages(file_path) print("✓ All Chinese messages replaced with English")