Trendy Post MCP

""" Post generator module for creating trending Xiaohongshu-style posts. """ import os import json import sys from typing import Dict, List, Any, Optional from datetime import datetime import random import logging from dotenv import load_dotenv from zai import ZhipuAiClient # Load environment variables load_dotenv() # Initialize OpenAI client LLM_API_KEY = os.getenv("ZHIPU_API_KEY") # Initialize logger logger = logging.getLogger(__name__) class PostGenerator: """ Generates trending Xiaohongshu-style posts based on image analysis and extracted text. """ def __init__(self): """Initialize the post generator.""" self.xiaohongshu_styles = [ "lifestyle", "fashion", "beauty", "food", "travel", "fitness", "home decor", "snowboarding", "bouldering", "archery", "AI", "news", "tech", "exhibitions", "concerts", "plays", "films", "series" ] self.emoji_sets = { "lifestyle": ["✨", "💫", "🌈", "💖", "🥰", "🌟", "🌱"], "fashion": ["👗", "👠", "👜", "💅", "👒", "✨", "💎"], "beauty": ["💄", "💋", "✨", "💆‍♀️", "💅", "🧴", "💫"], "food": ["🍜", "🍣", "🍰", "☕", "🍷", "🥂", "😋"], "travel": ["✈️", "🌍", "🏝️", "🗺️", "🧳", "📸", "🌅"], "fitness": ["💪", "🏃‍♀️", "🧘‍♀️", "🥗", "🥤", "🌱", "✨"], "home decor": ["🏠", "🪴", "🛋️", "✨", "🕯️", "🖼️", "💫"], "snowboarding": ["🏂", "❄️", "🏔️", "🌨️", "🎿", "🥶", "🔥"], "bouldering": ["🧗‍♀️", "🧗‍♂️", "💪", "🪨", "🧠", "🤸‍♀️", "✨"], "archery": ["🏹", "🎯", "🔄", "💯", "🧘‍♀️", "🏆", "✨"], "AI": ["🤖", "💻", "🧠", "✨", "🔮", "📊", "🚀"], "news": ["🌍", "🤝", "📜", "🏛️", "🔄", "📰", "🌐"], "exhibitions": ["🖼️", "🏛️", "🎨", "✨", "📸", "🖌️", "👁️"], "concerts": ["🎵", "🎤", "🎸", "🥁", "🎧", "✨", "🔥"], "plays": ["🎭", "🎬", "👥", "🎪", "🎟️", "✨", "👏"], "films": ["🎬", "🍿", "🎞️", "🎭", "🎥", "🎦", "✨"], "series": ["📺", "🍿", "🎬", "📱", "🎭", "🎞️", "✨"] } def _call_llm(self, prompt: str, system_prompt: str = "You are a helpful assistant.", max_tokens: int = 4096) -> str: """ Call the LLM to generate text. Args: prompt: The prompt to send to the LLM max_tokens: Maximum tokens to generate Returns: Generated text """ client = ZhipuAiClient(api_key=LLM_API_KEY) # 请填写您自己的 API Key messages = [ {"role": "system", "content": system_prompt}, {"role": "user", "content": prompt} ] print(messages) response = client.chat.completions.create( model="glm-4.5", messages=messages, thinking={ "type": "enabled", # 启用深度思考模式 }, max_tokens=max_tokens, # 最大输出tokens temperature=0.7 # 控制输出的随机性 ) print(response.choices[0].message.content) return response.choices[0].message.content def generate_post(self, image_data: Dict[str, Any], user_query: str = "") -> Dict[str, Any]: """ Generate a trending Xiaohongshu post based on image analysis. Args: image_data: Image analysis data user_query: Optional user query to guide post generation Returns: Generated post data """ # Extract relevant information from image data extracted_text = image_data.get("text", "") text_blocks = image_data.get("text_blocks", []) image_analysis = image_data.get("analysis", {}) # Determine the most appropriate style based on image content and user query style = self._determine_style(extracted_text, user_query) # Generate the post content content = self._generate_content(extracted_text, text_blocks, image_analysis, style, user_query) # Generate title based on content title = self.generate_title(content, style, user_query) # Generate hashtags hashtags = self._generate_hashtags(content, style, user_query) return { "title": title, "content": content, "hashtags": hashtags, "style": style } def _determine_style(self, text: str, user_query: str = "") -> str: """ Determine the most appropriate Xiaohongshu style based on image content using LLM. Args: text: Extracted text from the image image_analysis: Image analysis data user_query: Optional user query to guide style determination Returns: Style category string """ # Use LLM to determine the style based on the extracted text and image analysis prompt = f"""Based on the following information, determine the most appropriate style category for a Xiaohongshu post: Extracted text from image: {text} User query: {user_query} Please analyze this content and determine which of the following Xiaohongshu post styles would be most appropriate: {', '.join(self.xiaohongshu_styles)} Return only the style name, nothing else. """ try: # Call LLM to determine style response = self._call_llm(prompt) # Clean and validate the response suggested_style = response.strip().lower() # # Check if the suggested style is in our list of styles # if suggested_style in self.xiaohongshu_styles: # return suggested_style # # If not an exact match, try to find the closest match # for style in self.xiaohongshu_styles: # if style in suggested_style: # return style # Default to a random style if no match # return random.choice(self.xiaohongshu_styles) return suggested_style except Exception as e: logger.error(f"Error determining style with LLM: {str(e)}") # Fall back to random style if LLM fails return "lifestyle" def _generate_hashtags(self, content: str, style: str, user_query: str = "") -> List[str]: """ Generate hashtags for the post using LLM. Args: content: Generated post content style: Post style user_query: Optional user query to guide hashtag generation Returns: List of hashtags """ # prompt = f""" # Generate 5-8 trending Xiaohongshu hashtags for a post with the following: # Post content: {content[:300]} # Truncate to avoid token limits # Post style: {style} # User query: {user_query} # The hashtags should be relevant, trendy, and in the style of Xiaohongshu (小红书). # Return only the hashtags as a comma-separated list, without the # symbol. # """ prompt = f"""# Role: 小红书标签优化专家 ## 任务 基于用户提供的爆款文章正文，生成高转化率的小红书标签组合（Tags），需同时满足**搜索流量提升**和**算法推荐**双目标。 ### 标签生成策略（三层矩阵） 1. **流量池钥匙（占30%）** - 选择2-3个百万级泛流量词，覆盖基础用户池 - 要求：从当前平台热门标签中匹配（参考实时热搜词） - 示例：`#程序员` `#AI工具` `#效率提升` 2. **精准狙击器（占50%）** - 生成3-4个垂直领域标签，锁定细分人群需求 - 要求：结合正文关键词+行业高转化词（如技术类用`#独立开发者`，美妆类用`#黄黑皮天菜`） - 示例：`#初创公司技术栈` `#全栈开发` `#低成本创业` 3. **长尾钩子（占20%）** - 创建1-2个蓝海长尾词，避开头部竞争 - 要求： ▪️ 包含「解决方案+人群/场景」结构（例：`#学生党平价开发工具`） ▪️ 搜索量/内容量比值＞5（通过工具检测） ### 核心规则 ⚠️ **强制条款** - 标签总数：**严格控制在5-8个**（超出触发限流） - 排序逻辑：按「泛流量→垂类词→长尾词」顺序排列（前3位必须含大热词） - 敏感词规避：用「零克查词」检测，替换灰色词（如`#免费`→`#同价位更狠`） 🚫 **绝对禁忌** × 禁用重复标签（如同时用`#技术栈`和`#开发工具`） × 禁用失效标签（参考平台每月公示的「过时标签库」） × 禁用纯英文标签（中文标签曝光率高42%） ### 高阶技巧 1. **热点截流** - 若正文含热点关键词（如`AI`），添加带🔥图标的标签（例：`#AIGC工具🔥`） 2. **跨屏引流** - 同步抖音/微博热搜词（如`#多巴胺编程`），在标签和正文各出现3次 3. **养号策略** - 新账号前5篇笔记固定使用相同核心标签（例：技术类必带`#技术栈`+`#效率翻倍`） ### 输出格式 - 请直接返回你生成的标签，多个标签之间用英文逗号分隔，不要添加任何其他内容。 接下来，请你使用以下文本内容、风格、和用户对话内容，生成一个符合小红书风格的标签组合： 文本内容：{content[:300]} 风格：{style} 用户对话内容：{user_query} """ try: # Call LLM to generate hashtags response = self._call_llm(prompt) # Process the response hashtags = [tag.strip() for tag in response.split(',')] # Filter out empty tags and limit to 8 max hashtags = [tag for tag in hashtags if tag][:8] return hashtags except Exception as e: logger.error(f"Error generating hashtags with LLM: {str(e)}") # Return some generic hashtags if LLM fails return ["小红书", "分享", "推荐", style] def generate_title(self, content: str, style: str, user_query: str = "") -> str: """ Generate a catchy title for the post using LLM. Args: content: Post content style: Post style user_query: Optional user query to guide title generation Returns: Generated title """ # prompt = f""" # Generate a catchy, attention-grabbing Xiaohongshu (小红书) post title based on: # Post content: {content[:300]} # Post style: {style} # User query: {user_query} # The title should be concise (under 30 characters), engaging, and in the style of popular Xiaohongshu posts. # Return only the title, nothing else. # """ prompt = f"""**标题公式**： - 采用「人群+痛点/利益点+解决方案」结构 - 融合数字/悬念词/感叹词（如：程序员必看！3天提效200%的AI工具链💥） - 参考句式： * 震惊体："我竟然用XX天实现了XX！" * 数字体："XX个技巧让XX效率翻倍" * 反差体："别再XX了！这个方法YYY更有效" 请你根据以下文本内容、风格、和用户对话内容，生成一个符合小红书风格的标题： 文本内容：{content} 风格：{style} 用户对话内容：{user_query} 请直接返回你生成的标题，不要添加任何其他内容。 """ try: # Call LLM to generate title response = self._call_llm(prompt) # Clean the response title = response.strip() # # Ensure it's not too long # if len(title) > 50: # Allow some flexibility but still keep it concise # title = title[:47] + "..." return title except Exception as e: logger.error(f"Error generating title with LLM: {str(e)}") # Generate a simple title if LLM fails return f"我的{style}分享" # "My {style} sharing" def _generate_content( self, extracted_text: str, text_blocks: List[Dict[str, Any]], image_analysis: Dict[str, Any], post_style: str, user_query: str = "" ) -> str: """ Generate post content using LLM. Args: extracted_text: Text extracted from the image text_blocks: Structured text blocks from OCR image_analysis: Image analysis data post_style: Determined post style Returns: Generated post content """ try: # Prepare the prompt prompt = self._create_prompt(extracted_text, text_blocks, image_analysis, post_style, user_query) role_prompt = """# 角色设定 你是一名小红书爆款内容创作专家，擅长将普通话题转化为具有病毒式传播力的完整文章。你的核心能力包括： 1. 情绪钩子设计：运用“夸张情绪词+反常识观点”引爆好奇心（如“太可怕了”“谁懂啊”“绝绝子”） 2. 二极管结构法：通过“痛点刺激-解决方案-逆天效果”三段式框架制造爽感 3. 平台化表达：口语化行文+高频emoji+20字内短段落适配移动端阅读 # 爆款文章生成公式 ▶ 标题公式（保留原规则升级版） `[情绪词] + [反常识结果] + [人群标签] + [emoji]` ✅ 示例： > “太可怕了！用ChatGPT写文案月入5w+，打工人逆袭指南💥” ▶ 正文结构公式 1. 段落小标题（简洁明了） 2. 情绪冲击开头（引发共鸣） - 痛点场景故事（50字内）+ 夸张情绪词 > *例：“谁懂啊！熬夜写的文案0点赞，同事用AI 3秒收割10w流量！！”* 3. 颠覆认知转折（制造反差） - “直到我发现...” + 反常识方法（突出简单/速成） 4. 干货步骤拆解（实用价值） - 分步骤+emoji图标 + 具体案例（如“实测7天涨粉5千”） 5. 高潮金句点题（刺激传播） - 用“记住：...”句式+争议性观点（例：“不会用AI的文案人，终将被淘汰！”） 6. 互动钩子结尾（引导行动） - “评论区扣【666】领指令库” / “戳合集看100个变现案例” # 创作规则 1. 内容长度：正文300-500字，分3-5段，并配上段落小标题，每段≤3行 2. 关键词植入：从用户提供的关键词库选3个自然融入（如“吐血整理”“手残党必备”） 3. 平台特性： - 每段尽量有段落小标题，并在段落标题和正文中插入1-2个🔥⭐💡类高亮emoji - 关键信息用“！”标点强化情绪 4. 禁忌： × 避免长段落（每段≤3行） × 禁止直接出现"小红书"、"粉丝"等平台词 × 禁用复杂标签（后续单独处理） """ response = self._call_llm(prompt, system_prompt=role_prompt, max_tokens=8192) # print(response) return response except Exception as e: # Fallback content in case of API error print(f"Error generating content: {e}") return self._generate_fallback_content(extracted_text, post_style) def _create_prompt( self, extracted_text: str, text_blocks: List[Dict[str, Any]], image_analysis: Dict[str, Any], post_style: str, user_query: str = "" ) -> str: """ Create a prompt for the LLM to generate post content. Args: extracted_text: Text extracted from the image text_blocks: Structured text blocks from OCR image_analysis: Image analysis data post_style: Determined post style Returns: Prompt string """ # Extract relevant image information dimensions = image_analysis.get("dimensions", {}) color_info = image_analysis.get("color_info", {}) # Image details: # - Dimensions: {dimensions.get('width', 'unknown')}x{dimensions.get('height', 'unknown')} # - Brightness: {'Bright' if color_info.get('is_bright', False) else 'Dark'} prompt = f"""请你根据以下文本内容、风格、和用户对话内容，生成一个小红书爆款文章风格的推文： 文本内容：{extracted_text} 风格：{post_style} 用户对话内容：{user_query} 请注意：善用情绪钩子、平台化表达。多分段落，并添加emoji。请你直接返回你生成的推文，不要添加任何其他内容。 """ return prompt def _generate_fallback_content(self, extracted_text: str, post_style: str) -> str: """ Generate fallback content when API call fails. Args: extracted_text: Text extracted from the image post_style: Determined post style Returns: Fallback content """ # Extract a few words from the text to personalize the fallback words = extracted_text.split() sample_text = " ".join(words[:10]) if len(words) > 10 else extracted_text templates = [ f"今天发现了一个超棒的{post_style}好物！{sample_text}... 真的很推荐大家尝试一下～", f"分享一下我最近超爱的{post_style}心得！{sample_text}... 希望对你也有帮助～", f"偶然间发现的{post_style}宝藏！{sample_text}... 真的是相见恨晚！" ] return random.choice(templates) def _get_emojis(self, post_style: str, count: int = 5) -> List[str]: """ Get relevant emojis for the post. Args: post_style: Post style category count: Maximum number of emojis to include Returns: List of emoji strings """ style_emojis = self.emoji_sets.get(post_style, ["✨", "💫", "🌈", "💖", "🥰", "🌟", "🌱"]) # Select random emojis from the style category if style_emojis: # Ensure we don't try to select more emojis than available available_count = min(count, len(style_emojis)) selected_emojis = random.sample(style_emojis, available_count) else: selected_emojis = ["✨", "💫"] return selected_emojis def _format_post(self, content: str, hashtags: List[str], emojis: List[str]) -> str: """ Format the final post with content, hashtags, and emojis. Args: content: Generated post content hashtags: Selected hashtags emojis: Selected emojis Returns: Formatted post string """ # Add random emojis throughout the content paragraphs = content.split("\n\n") formatted_paragraphs = [] for i, paragraph in enumerate(paragraphs): if paragraph.strip(): # Add an emoji at the beginning or end of some paragraphs if random.random() > 0.5 and emojis: emoji = random.choice(emojis) if random.random() > 0.5: paragraph = f"{emoji} {paragraph}" else: paragraph = f"{paragraph} {emoji}" formatted_paragraphs.append(paragraph) # Join paragraphs formatted_content = "\n\n".join(formatted_paragraphs) # Add hashtags at the end hashtag_section = " ".join(hashtags) # Final post final_post = f"{formatted_content}\n\n{hashtag_section}" return final_post # Example usage if __name__ == "__main__": generator = PostGenerator() # Sample image data sample_data = { "text": "今天去了新开的咖啡店，环境超级好，咖啡也很香浓", "text_blocks": [], "language": "zh", "analysis": { "dimensions": {"width": 1080, "height": 1920, "aspect_ratio": 0.5625}, "color_info": {"is_bright": True, "dominant_colors": [[240, 240, 240], [200, 180, 160]]} } } post = generator.generate_post(sample_data) print(json.dumps(post, ensure_ascii=False, indent=2))