content_gen.py

"""
content_gen.py — LLM Layer (Local Copy)
Generates a structured manifest from a topic prompt.
Updated to use DashScope (Alibaba Cloud) with Qwen3.6-Plus.
"""
import json
import re
import os
from typing import Dict, Any
import requests
from datetime import datetime


class ContentGenerator:
    # API endpoints
    # Using the international endpoint as it successfully authenticated with the provided key
    DASHSCOPE_API_URL = "https://dashscope-intl.aliyuncs.com/compatible-mode/v1"
    
    def __init__(self, config: Dict[str, Any]):
        self.config = config
        # ENFORCE qwen3.6-plus to avoid 404 errors from outdated config values
        self.llm_model = "qwen3.6-plus"
        
        # DashScope API key provided by user
        self.dashscope_api_key = "sk-ws-H.HRRIYI.hUUL.MEUCIQDv1QpE1B2xqwrl2OUSt1o7XbluYkzzaW1sCkp_FMCrewIgSbClEsN--mLhn2JAWt5kyrmaX30grEhaAGcav2TeLS0"
        self.api_url = self.DASHSCOPE_API_URL
        self.scenes_per_video = config.get("scenes_per_video", 8)
        
        if not self.dashscope_api_key:
            raise ValueError("DashScope API key is not provided")

    def generate_manifest(self, topic: str) -> Dict[str, Any]:
        """
        Generate a manifest (scene list, caption, hashtags) from a topic prompt.
        
        Args:
            topic: User's topic prompt (e.g., "sunset photography tips")
        
        Returns:
            manifest dict with scenes, caption, hashtags
        """
        print(f"[content] Generating manifest for topic: '{topic}' using {self.llm_model}")
        
        # Build LLM prompt
        prompt = self._build_prompt(topic)
        
        # Call DashScope API
        manifest_text = self._call_llm(prompt)
        
        # Parse JSON from response
        manifest = self._parse_manifest(manifest_text, topic)
        
        print(f"[content] Manifest ready — {len(manifest['scenes'])} scenes")
        return manifest

    def _build_prompt(self, topic: str) -> str:
        """Build the LLM prompt to generate a manifest."""
        return f"""You are a viral TikTok/Reels content expert. Generate a structured JSON manifest for a short-form video.

Topic: {topic}

Generate exactly {self.scenes_per_video} scenes. Each scene should:
- Have a memorable label (3-5 words, can start with ~)
- Have an aesthetic image_query (for Pinterest/Google Images search)

Also provide:
- A catchy title
- An engaging caption (hooks viewers to comment/share)
- 4-5 trending hashtags

Respond with ONLY valid JSON, no markdown or extra text:

{{
  "title": "...",
  "scenes": [
    {{"label": "...", "image_query": "..."}},
    ...
  ],
  "caption": "...",
  "hashtags": ["#...", "#...", ...]
}}"""

    def _call_llm(self, prompt: str) -> str:
        """Call the LLM via DashScope OpenAI-compatible API."""
        headers = {
            "Authorization": f"Bearer {self.dashscope_api_key}",
            "Content-Type": "application/json",
        }
        
        payload = {
            "model": self.llm_model,
            "messages": [
                {
                    "role": "user",
                    "content": prompt
                }
            ],
            "temperature": 0.7,
            "max_tokens": 2000,
        }
        
        try:
            response = requests.post(
                f"{self.api_url}/chat/completions",
                headers=headers,
                json=payload,
                timeout=180
            )
            response.raise_for_status()
            data = response.json()
            
            # Check for errors in response
            if "error" in data:
                raise Exception(f"API error: {data['error']}")
            
            content = data.get("choices", [{}])[0].get("message", {}).get("content")
            if not content:
                raise Exception("Empty response from LLM")
            
            # Remove potential markdown code blocks if the model included them
            content = re.sub(r'^```json\s*', '', content, flags=re.MULTILINE)
            content = re.sub(r'\s*```$', '', content, flags=re.MULTILINE)
            
            return content.strip()
        except Exception as e:
            print(f"[content] ERROR calling LLM: {e}")
            raise

    def _parse_manifest(self, text: str, topic: str) -> Dict[str, Any]:
        """Parse and validate the JSON manifest from LLM response."""
        if not text:
            print(f"[content] ERROR: Empty response from LLM, using fallback manifest")
            return self._create_fallback_manifest(topic)
        
        try:
            # Try to extract JSON from the response
            manifest = json.loads(text)
        except json.JSONDecodeError:
            # Try to find JSON block in case there's still extra text
            json_match = re.search(r'\{.*\}', text, re.DOTALL)
            if json_match:
                try:
                    manifest = json.loads(json_match.group())
                except json.JSONDecodeError:
                    print(f"[content] WARNING: Could not parse LLM JSON, using fallback manifest")
                    return self._create_fallback_manifest(topic)
            else:
                print(f"[content] WARNING: No JSON found in response, using fallback manifest")
                return self._create_fallback_manifest(topic)
        
        # Validate structure
        required_keys = ["title", "scenes", "caption", "hashtags"]
        for key in required_keys:
            if key not in manifest:
                print(f"[content] WARNING: Missing key '{key}' in manifest")
                fallback = self._create_fallback_manifest(topic)
                manifest[key] = fallback.get(key)
        
        # Ensure exactly scenes_per_video scenes
        if "scenes" in manifest and isinstance(manifest["scenes"], list):
            if len(manifest["scenes"]) != self.scenes_per_video:
                print(f"[content] WARNING: Expected {self.scenes_per_video} scenes, got {len(manifest['scenes'])}")
                if len(manifest['scenes']) > self.scenes_per_video:
                    manifest["scenes"] = manifest["scenes"][:self.scenes_per_video]
                else:
                    while len(manifest["scenes"]) < self.scenes_per_video:
                        manifest["scenes"].append({
                            "label": f"Bonus Tip {len(manifest['scenes'])+1}",
                            "image_query": topic
                        })
        
        # Ensure each scene has label and image_query
        if "scenes" in manifest and isinstance(manifest["scenes"], list):
            for i, scene in enumerate(manifest["scenes"]):
                if not isinstance(scene, dict):
                    manifest["scenes"][i] = {"label": str(scene), "image_query": topic}
                    continue
                if "label" not in scene:
                    scene["label"] = f"Scene {i+1}"
                if "image_query" not in scene:
                    scene["image_query"] = topic
        
        # Remove emojis from caption and hashtags
        if "caption" in manifest and isinstance(manifest["caption"], str):
            manifest["caption"] = self._remove_emojis(manifest["caption"])
        if "hashtags" in manifest and isinstance(manifest["hashtags"], list):
            manifest["hashtags"] = [self._remove_emojis(str(tag)) for tag in manifest["hashtags"]]
        
        # Add timestamp
        manifest["timestamp"] = datetime.now().isoformat()
        
        return manifest

    def _create_fallback_manifest(self, topic: str) -> Dict[str, Any]:
        """Create a basic fallback manifest if LLM fails."""
        return {
            "title": topic.title(),
            "scenes": [
                {"label": f"Scene {i+1}", "image_query": topic}
                for i in range(self.scenes_per_video)
            ],
            "caption": "Which one hits hardest?",
            "hashtags": ["#relatable", "#fyp", "#trending"],
            "timestamp": datetime.now().isoformat()
        }

    def _remove_emojis(self, text: str) -> str:
        """Remove all emojis from text, keeping only ASCII characters and common punctuation."""
        emoji_pattern = re.compile(
            "["
            "\U0001F600-\U0001F64F"  # Emoticons
            "\U0001F300-\U0001F5FF"  # Symbols & pictographs
            "\U0001F680-\U0001F6FF"  # Transport & map symbols
            "\U0001F1E0-\U0001F1FF"  # Flags (iOS)
            "\U00002702-\U000027B0"
            "\U000024C2-\U0001F251"
            "\U0001f926-\U0001f937"
            "\U00010000-\U0010ffff"
            "\u2640-\u2642"
            "\u2600-\u2B55"
            "\u200d"
            "\u23cf"
            "\u23e9"
            "\u231a"
            "\ufe0f"  # Dingbats
            "\u3030"
            "]+", re.UNICODE
        )
        text = emoji_pattern.sub(' ', text)
        text = re.sub(r'\s+', ' ', text).strip()
        return text