feat: podcast demo mode with ALWRITY_ENABLED_FEATURES support
- Add ALWRITY_ENABLED_FEATURES env var for feature gating - Podcast-only mode: skip LLM bootstrap, scheduler, persona services - Enhance video generation prompt with scene context, analysis, narration - Add voice cloning support via custom_voice_id in WaveSpeed - Add text-to-speech for research results (browser speechSynthesis) - Fix render queue to sync images from script phase - Add WaveSpeed LLM pricing (gpt-oss-120b) - Fix podcast bible generation error handling - Refactor RouterManager for feature-based router loading
This commit is contained in:
ajaysi
@@ -6,6 +6,7 @@ Centralized constants and directory configuration for podcast module.
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Literal
|
||||
from loguru import logger
|
||||
from services.story_writer.audio_generation_service import StoryAudioGenerationService
|
||||
|
||||
# Directory paths
|
||||
@@ -45,11 +46,14 @@ def get_podcast_media_dir(
|
||||
}[media_type]
|
||||
|
||||
if user_id:
|
||||
tenant_media_dir = ROOT_DIR / "workspace" / f"workspace_{_sanitize_user_id(user_id)}" / "media" / media_subdir
|
||||
sanitized = _sanitize_user_id(user_id)
|
||||
tenant_media_dir = ROOT_DIR / "workspace" / f"workspace_{sanitized}" / "media" / media_subdir
|
||||
resolved_dir = tenant_media_dir.resolve()
|
||||
else:
|
||||
resolved_dir = (DATA_MEDIA_DIR / media_subdir).resolve()
|
||||
|
||||
logger.debug(f"[Podcast] get_podcast_media_dir: type={media_type}, user_id={user_id}, sanitized={user_id and _sanitize_user_id(user_id)}, resolved={resolved_dir}")
|
||||
|
||||
if ensure_exists:
|
||||
resolved_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
@@ -61,7 +65,9 @@ def get_podcast_media_read_dirs(media_type: MediaType, user_id: str | None = Non
|
||||
dirs: list[Path] = []
|
||||
if user_id:
|
||||
dirs.append(get_podcast_media_dir(media_type, user_id))
|
||||
logger.debug(f"[Podcast] get_podcast_media_read_dirs: added user dir for {user_id}")
|
||||
dirs.append(get_podcast_media_dir(media_type, None))
|
||||
logger.debug(f"[Podcast] get_podcast_media_read_dirs: dirs={dirs}")
|
||||
return dirs
|
||||
|
||||
|
||||
|
||||
@@ -5,10 +5,11 @@ Analysis endpoint for podcast ideas.
|
||||
"""
|
||||
|
||||
from fastapi import APIRouter, Depends, HTTPException
|
||||
from typing import Dict, Any
|
||||
from typing import Dict, Any, Optional, List
|
||||
import json
|
||||
import uuid
|
||||
from sqlalchemy.orm import Session
|
||||
from pydantic import BaseModel
|
||||
|
||||
from services.database import get_db
|
||||
from middleware.auth_middleware import get_current_user
|
||||
@@ -258,6 +259,10 @@ Return JSON with:
|
||||
- top_keywords: 5 podcast-relevant keywords/phrases
|
||||
- suggested_outlines: 2 items, each with title (<=60 chars) and 4-6 short segments (bullet-friendly, factual)
|
||||
- title_suggestions: 3 concise episode titles
|
||||
- episode_hook: one compelling 15-30 second opening hook/angle that grabs attention
|
||||
- key_takeaways: 3-5 actionable insights listeners will learn
|
||||
- guest_talking_points: (if guest included) 3-4 suggested questions/angles for guest interview
|
||||
- listener_cta: one clear call-to-action for listeners
|
||||
- research_queries: array of {{"query": "string", "rationale": "string"}}
|
||||
- exa_suggested_config: suggested Exa search options with:
|
||||
- exa_search_type: "auto" | "neural" | "keyword"
|
||||
@@ -271,7 +276,10 @@ Return JSON with:
|
||||
Requirements:
|
||||
- Keep language factual, actionable, and suited for spoken audio.
|
||||
- Avoid narrative fiction tone.
|
||||
- Prefer 2024-2025 context.
|
||||
- For research queries: Mix of time-sensitive and evergreen queries:
|
||||
- 2-3 queries should focus on latest 2025-2026 developments, trends, and data (use year in query)
|
||||
- 2-3 queries should be evergreen/fundamental (concepts, definitions, best practices, proven strategies) - do NOT include years in these
|
||||
- Today's date is April 2026.
|
||||
"""
|
||||
|
||||
try:
|
||||
@@ -305,6 +313,10 @@ Requirements:
|
||||
top_keywords = data.get("top_keywords") or []
|
||||
suggested_outlines = data.get("suggested_outlines") or []
|
||||
title_suggestions = data.get("title_suggestions") or []
|
||||
episode_hook = data.get("episode_hook") or ""
|
||||
key_takeaways = data.get("key_takeaways") or []
|
||||
guest_talking_points = data.get("guest_talking_points") or []
|
||||
listener_cta = data.get("listener_cta") or ""
|
||||
research_queries = data.get("research_queries") or []
|
||||
exa_suggested_config = data.get("exa_suggested_config") or None
|
||||
|
||||
@@ -314,6 +326,10 @@ Requirements:
|
||||
top_keywords=top_keywords,
|
||||
suggested_outlines=suggested_outlines,
|
||||
title_suggestions=title_suggestions,
|
||||
episode_hook=episode_hook,
|
||||
key_takeaways=key_takeaways,
|
||||
guest_talking_points=guest_talking_points,
|
||||
listener_cta=listener_cta,
|
||||
research_queries=research_queries,
|
||||
exa_suggested_config=exa_suggested_config,
|
||||
bible=bible_obj.model_dump() if bible_obj else None,
|
||||
@@ -321,3 +337,106 @@ Requirements:
|
||||
avatar_prompt=final_avatar_prompt,
|
||||
)
|
||||
|
||||
|
||||
class RegenerateQueriesRequest(BaseModel):
|
||||
idea: str
|
||||
feedback: str
|
||||
existing_analysis: Optional[Dict[str, Any]] = None
|
||||
bible: Optional[Dict[str, Any]] = None
|
||||
|
||||
|
||||
class RegenerateQueriesResponse(BaseModel):
|
||||
research_queries: List[Dict[str, str]]
|
||||
|
||||
|
||||
@router.post("/regenerate-queries", response_model=RegenerateQueriesResponse)
|
||||
async def regenerate_research_queries(
|
||||
request: RegenerateQueriesRequest,
|
||||
current_user: Dict[str, Any] = Depends(get_current_user),
|
||||
):
|
||||
"""
|
||||
Regenerate research queries based on user feedback and existing analysis.
|
||||
"""
|
||||
user_id = require_authenticated_user(current_user)
|
||||
|
||||
# Build context from existing analysis
|
||||
idea = request.idea
|
||||
feedback = request.feedback
|
||||
|
||||
# Get topic, keywords, audience from existing analysis if provided
|
||||
topic = idea
|
||||
keywords = ""
|
||||
audience = ""
|
||||
if request.existing_analysis:
|
||||
topic = request.existing_analysis.get("title_suggestions", [idea])[0] if request.existing_analysis.get("title_suggestions") else idea
|
||||
keywords = ", ".join(request.existing_analysis.get("top_keywords", [])[:5])
|
||||
audience = request.existing_analysis.get("audience", "")
|
||||
|
||||
# Serialize Bible context if provided
|
||||
bible_context = ""
|
||||
if request.bible:
|
||||
try:
|
||||
bible_service = PodcastBibleService()
|
||||
from models.podcast_bible_models import PodcastBible
|
||||
bible_data = PodcastBible(**request.bible)
|
||||
bible_context = bible_service.serialize_bible(bible_data)
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to serialize bible for query regeneration: {e}")
|
||||
|
||||
prompt = f"""
|
||||
You are a research strategist for podcast content. Given a podcast idea, existing analysis, and user feedback,
|
||||
generate 7 new research queries that address the user's specific needs.
|
||||
|
||||
{f"USER FEEDBACK: {feedback}" if feedback else ""}
|
||||
|
||||
{f"EXISTING ANALYSIS CONTEXT:\n- Topic: {topic}\n- Keywords: {keywords}\n- Audience: {audience}\n" if request.existing_analysis else ""}
|
||||
{f"PODCAST BIBLE CONTEXT:\n{bible_context}\n" if bible_context else ""}
|
||||
|
||||
Podcast Idea: "{idea}"
|
||||
|
||||
TASK:
|
||||
Generate exactly 7 research queries that:
|
||||
1. Incorporate the user's feedback direction
|
||||
2. Build on the existing analysis context
|
||||
3. Mix of time-sensitive (2025-2026) and evergreen topics
|
||||
4. Are highly specific to the podcast topic
|
||||
|
||||
Return JSON with:
|
||||
- research_queries: array of {{"query": "string", "rationale": "string"}}
|
||||
|
||||
Requirements:
|
||||
- At least 2-3 queries should focus on latest 2025-2026 developments (include year in query)
|
||||
- At least 2-3 queries should be evergreen (concepts, definitions, best practices - NO year)
|
||||
- Queries should be specific and actionable, not generic
|
||||
"""
|
||||
|
||||
try:
|
||||
from services.llm_providers.main_text_generation import llm_text_gen
|
||||
|
||||
raw = llm_text_gen(
|
||||
prompt=prompt,
|
||||
user_id=user_id,
|
||||
json_struct={"research_queries": [{"query": "string", "rationale": "string"}]},
|
||||
preferred_provider=None,
|
||||
flow_type="premium_tool",
|
||||
)
|
||||
|
||||
# Parse response
|
||||
if isinstance(raw, dict):
|
||||
queries = raw.get("research_queries", [])
|
||||
else:
|
||||
# Try to parse as JSON
|
||||
try:
|
||||
parsed = json.loads(raw) if isinstance(raw, str) else raw
|
||||
queries = parsed.get("research_queries", []) if isinstance(parsed, dict) else []
|
||||
except:
|
||||
queries = []
|
||||
|
||||
return RegenerateQueriesResponse(research_queries=queries[:7])
|
||||
|
||||
except HTTPException:
|
||||
raise
|
||||
except Exception as exc:
|
||||
logger.error(f"[Regenerate Queries] Failed for user {user_id}: {exc}")
|
||||
raise HTTPException(status_code=500, detail=f"Regenerate queries failed: {exc}")
|
||||
|
||||
|
||||
@@ -126,12 +126,14 @@ async def generate_podcast_audio(
|
||||
|
||||
try:
|
||||
audio_service = get_podcast_audio_service(user_id)
|
||||
logger.warning(f"[Podcast] Generating audio with service dir: {audio_service.output_dir}")
|
||||
result: StoryAudioResult = audio_service.generate_ai_audio(
|
||||
scene_number=0,
|
||||
scene_title=request.scene_title,
|
||||
text=request.text.strip(),
|
||||
user_id=user_id,
|
||||
voice_id=request.voice_id or "Wise_Woman",
|
||||
custom_voice_id=request.custom_voice_id,
|
||||
speed=request.speed or 1.0, # Normal speed (was 0.9, but too slow - causing duration issues)
|
||||
volume=request.volume or 1.0,
|
||||
pitch=request.pitch or 0.0, # Normal pitch (0.0 = neutral)
|
||||
@@ -149,6 +151,8 @@ async def generate_podcast_audio(
|
||||
if result.get("audio_url") and "/api/story/audio/" in result.get("audio_url", ""):
|
||||
audio_filename = result.get("audio_filename", "")
|
||||
result["audio_url"] = f"/api/podcast/audio/{audio_filename}"
|
||||
|
||||
logger.warning(f"[Podcast] Audio generated - path: {result.get('audio_path')}, url: {result.get('audio_url')}")
|
||||
except Exception as exc:
|
||||
raise HTTPException(status_code=500, detail=f"Audio generation failed: {exc}")
|
||||
|
||||
@@ -387,7 +391,9 @@ async def serve_podcast_audio(
|
||||
raise HTTPException(status_code=400, detail="Invalid filename")
|
||||
|
||||
user_id = require_authenticated_user(current_user)
|
||||
logger.warning(f"[Podcast] serve_podcast_audio called: user_id={user_id}, filename={filename}")
|
||||
audio_path = _resolve_podcast_media_file(filename, "audio", user_id)
|
||||
logger.warning(f"[Podcast] Resolved audio path: {audio_path}")
|
||||
|
||||
return FileResponse(audio_path, media_type="audio/mpeg")
|
||||
|
||||
|
||||
@@ -104,6 +104,16 @@ async def generate_podcast_scene_image(
|
||||
# Otherwise, generate from scratch with podcast-optimized prompt
|
||||
image_prompt = "" # Initialize prompt variable
|
||||
|
||||
# Emotion to lighting mapping for visual tone
|
||||
emotion_lighting = {
|
||||
"happy": "warm, bright lighting, cheerful atmosphere",
|
||||
"excited": "dynamic, energetic lighting with highlights",
|
||||
"serious": "professional, balanced lighting, authoritative feel",
|
||||
"curious": "soft, inviting lighting, thoughtful atmosphere",
|
||||
"confident": "strong, dramatic lighting, authoritative look",
|
||||
"neutral": "professional, balanced lighting"
|
||||
}
|
||||
|
||||
if base_avatar_bytes:
|
||||
# Use Ideogram Character API for consistent character generation
|
||||
# Use custom prompt if provided, otherwise build scene-specific prompt
|
||||
@@ -127,6 +137,28 @@ async def generate_podcast_scene_image(
|
||||
if bible_obj.host.look:
|
||||
prompt_parts.append(f"Host Look: {bible_obj.host.look}")
|
||||
|
||||
# Scene emotion for visual tone
|
||||
emotion_lighting = {
|
||||
"happy": "warm, bright lighting, cheerful atmosphere",
|
||||
"excited": "dynamic, energetic lighting with highlights",
|
||||
"serious": "professional, balanced lighting, authoritative feel",
|
||||
"curious": "soft, inviting lighting, thoughtful atmosphere",
|
||||
"confident": "strong, dramatic lighting, authoritative look",
|
||||
"neutral": "professional, balanced lighting"
|
||||
}
|
||||
scene_emotion = request.scene_emotion
|
||||
if scene_emotion and scene_emotion in emotion_lighting:
|
||||
prompt_parts.append(emotion_lighting[scene_emotion])
|
||||
|
||||
# AI Analysis context for visual relevance
|
||||
if request.analysis:
|
||||
keywords = request.analysis.get("topKeywords", [])[:5]
|
||||
if keywords:
|
||||
prompt_parts.append(f"Keywords: {', '.join(keywords)}")
|
||||
audience = request.analysis.get("audience", "")
|
||||
if audience:
|
||||
prompt_parts.append(f"Target: {audience}")
|
||||
|
||||
# Scene content insights for visual context
|
||||
if request.scene_content:
|
||||
content_preview = request.scene_content[:200].replace("\n", " ").strip()
|
||||
@@ -139,6 +171,12 @@ async def generate_podcast_scene_image(
|
||||
visual_keywords.append("modern tech studio setting")
|
||||
if any(word in content_lower for word in ["business", "growth", "strategy", "market"]):
|
||||
visual_keywords.append("professional business studio")
|
||||
if any(word in content_lower for word in ["nature", "outdoor", "environment", "green"]):
|
||||
visual_keywords.append("natural outdoor setting")
|
||||
if any(word in content_lower for word in ["medical", "health", "wellness"]):
|
||||
visual_keywords.append("clean medical studio")
|
||||
if any(word in content_lower for word in ["education", "learning", "students"]):
|
||||
visual_keywords.append("classroom or educational setting")
|
||||
if visual_keywords:
|
||||
prompt_parts.append(", ".join(visual_keywords))
|
||||
|
||||
@@ -265,6 +303,19 @@ async def generate_podcast_scene_image(
|
||||
if request.scene_title:
|
||||
prompt_parts.append(f"Scene theme: {request.scene_title}")
|
||||
|
||||
# Scene emotion for visual tone (no avatar branch)
|
||||
if request.scene_emotion and request.scene_emotion in emotion_lighting:
|
||||
prompt_parts.append(emotion_lighting[request.scene_emotion])
|
||||
|
||||
# AI Analysis context (no avatar branch)
|
||||
if request.analysis:
|
||||
keywords = request.analysis.get("topKeywords", [])[:5]
|
||||
if keywords:
|
||||
prompt_parts.append(f"Keywords: {', '.join(keywords)}")
|
||||
audience = request.analysis.get("audience", "")
|
||||
if audience:
|
||||
prompt_parts.append(f"Target: {audience}")
|
||||
|
||||
# Content context for visual relevance
|
||||
if request.scene_content:
|
||||
content_preview = request.scene_content[:150].replace("\n", " ").strip()
|
||||
@@ -276,6 +327,12 @@ async def generate_podcast_scene_image(
|
||||
visual_keywords.append("modern technology aesthetic")
|
||||
if any(word in content_lower for word in ["business", "growth", "strategy", "market"]):
|
||||
visual_keywords.append("professional business environment")
|
||||
if any(word in content_lower for word in ["nature", "outdoor", "environment"]):
|
||||
visual_keywords.append("natural outdoor setting")
|
||||
if any(word in content_lower for word in ["medical", "health", "wellness"]):
|
||||
visual_keywords.append("clean medical studio")
|
||||
if any(word in content_lower for word in ["education", "learning", "students"]):
|
||||
visual_keywords.append("classroom or educational setting")
|
||||
if visual_keywords:
|
||||
prompt_parts.append(", ".join(visual_keywords))
|
||||
|
||||
@@ -379,6 +436,7 @@ async def generate_podcast_scene_image(
|
||||
provider=result.provider,
|
||||
model=result.model,
|
||||
cost=cost,
|
||||
image_prompt=image_prompt,
|
||||
)
|
||||
|
||||
except HTTPException:
|
||||
|
||||
@@ -27,7 +27,10 @@ async def create_project(
|
||||
db: Session = Depends(get_db),
|
||||
current_user: Dict[str, Any] = Depends(get_current_user),
|
||||
):
|
||||
"""Create a new podcast project."""
|
||||
"""Create a new podcast project.
|
||||
|
||||
If a project with the same idea already exists, return 409 conflict with existing project info.
|
||||
"""
|
||||
try:
|
||||
user_id = current_user.get("user_id") or current_user.get("id")
|
||||
if not user_id:
|
||||
@@ -40,6 +43,19 @@ async def create_project(
|
||||
if existing:
|
||||
raise HTTPException(status_code=400, detail="Project ID already exists")
|
||||
|
||||
# Check for duplicate idea (case-insensitive partial match)
|
||||
existing_idea = service.get_project_by_idea(user_id, request.idea)
|
||||
if existing_idea:
|
||||
raise HTTPException(
|
||||
status_code=409,
|
||||
detail={
|
||||
"message": "A project with similar idea already exists",
|
||||
"existing_project_id": existing_idea.project_id,
|
||||
"existing_idea": existing_idea.idea,
|
||||
"existing_status": existing_idea.status,
|
||||
}
|
||||
)
|
||||
|
||||
project = service.create_project(
|
||||
user_id=user_id,
|
||||
project_id=request.project_id,
|
||||
|
||||
@@ -8,6 +8,7 @@ from fastapi import APIRouter, Depends, HTTPException
|
||||
from typing import Dict, Any, List
|
||||
from types import SimpleNamespace
|
||||
import json
|
||||
import re
|
||||
|
||||
from middleware.auth_middleware import get_current_user
|
||||
from api.story_writer.utils.auth import require_authenticated_user
|
||||
@@ -36,10 +37,16 @@ async def podcast_research_exa(
|
||||
Uses Podcast Bible and Analysis context for hyper-personalization.
|
||||
"""
|
||||
user_id = require_authenticated_user(current_user)
|
||||
logger.warning(f"[Podcast Research] ========== REQUEST START ==========")
|
||||
logger.warning(f"[Podcast Research] User: {user_id}, Topic: {request.topic[:80]}...")
|
||||
logger.warning(f"[Podcast Research] Queries count: {len(request.queries) if request.queries else 0}")
|
||||
|
||||
|
||||
queries = [q.strip() for q in request.queries if q and q.strip()]
|
||||
if not queries:
|
||||
raise HTTPException(status_code=400, detail="At least one query is required for research.")
|
||||
|
||||
logger.warning(f"[Podcast Research] EXACT queries being sent to Exa: {queries}")
|
||||
|
||||
exa_cfg = request.exa_config or PodcastExaConfig()
|
||||
cfg = SimpleNamespace(
|
||||
@@ -52,6 +59,7 @@ async def podcast_research_exa(
|
||||
)
|
||||
|
||||
provider = ExaResearchProvider()
|
||||
logger.warning(f"[Podcast Research] Provider initialized, starting Exa search...")
|
||||
|
||||
# --- Context Building ---
|
||||
bible_service = PodcastBibleService()
|
||||
@@ -68,9 +76,16 @@ async def podcast_research_exa(
|
||||
if request.analysis:
|
||||
analysis_context = f"""
|
||||
PODCAST ANALYSIS CONTEXT:
|
||||
Audience: {request.analysis.get('audience', 'General')}
|
||||
========================
|
||||
Topic: {request.topic}
|
||||
Target Audience: {request.analysis.get('audience', 'General')}
|
||||
Content Type: {request.analysis.get('content_type', 'Informative')}
|
||||
Top Keywords: {', '.join(request.analysis.get('top_keywords', []))}
|
||||
|
||||
Episode Hook (Intro): {request.analysis.get('episode_hook', 'N/A')}
|
||||
Key Takeaways: {', '.join(request.analysis.get('key_takeaways', [])) or 'N/A'}
|
||||
Guest Talking Points: {', '.join(request.analysis.get('guest_talking_points', [])) or 'N/A'}
|
||||
Listener CTA: {request.analysis.get('listener_cta', 'N/A')}
|
||||
"""
|
||||
|
||||
# Exa search params
|
||||
@@ -84,6 +99,7 @@ Top Keywords: {', '.join(request.analysis.get('top_keywords', []))}
|
||||
|
||||
try:
|
||||
# 1. RUN EXA SEARCH
|
||||
logger.warning(f"[Podcast Research] Calling Exa search with topic: {request.topic[:100]}...")
|
||||
result = await provider.search(
|
||||
prompt=request.topic,
|
||||
topic=request.topic,
|
||||
@@ -92,8 +108,9 @@ Top Keywords: {', '.join(request.analysis.get('top_keywords', []))}
|
||||
config=cfg,
|
||||
user_id=user_id,
|
||||
)
|
||||
logger.warning(f"[Podcast Research] Exa search completed, got {len(result.get('sources', []))} sources")
|
||||
except Exception as exc:
|
||||
logger.error(f"[Podcast Exa Research] Search failed for user {user_id}: {exc}")
|
||||
logger.error(f"[Podcast Exa Research] Search failed for user {user_id}: {exc}", exc_info=True)
|
||||
raise HTTPException(status_code=500, detail=f"Exa research failed: {exc}")
|
||||
|
||||
# 2. EXTRACT INSIGHTS VIA LLM
|
||||
@@ -104,46 +121,77 @@ Top Keywords: {', '.join(request.analysis.get('top_keywords', []))}
|
||||
key_insights = []
|
||||
|
||||
if raw_content and sources:
|
||||
logger.info(f"[Podcast Research] Extracting insights from {len(sources)} sources for user {user_id}")
|
||||
logger.warning(f"[Podcast Research] Extracting insights from {len(sources)} sources for user {user_id}")
|
||||
|
||||
# Build list of research queries used for this search
|
||||
queries_used = ", ".join([f"Query {i+1}: {q}" for i, q in enumerate(queries)]) if queries else "No specific queries"
|
||||
|
||||
prompt = f"""
|
||||
You are an expert research analyst for a high-end podcast production team.
|
||||
Your task is to analyze the following research data and extract deep, actionable insights for a podcast episode.
|
||||
You are an expert research analyst and content strategist for a high-end podcast production team.
|
||||
Your task is to analyze the research data and extract deep, podcast-ready insights.
|
||||
|
||||
PODCAST CONTEXT:
|
||||
Topic: {request.topic}
|
||||
================
|
||||
Main Topic: {request.topic}
|
||||
|
||||
RESEARCH QUERIES USED:
|
||||
=====================
|
||||
{queries_used}
|
||||
|
||||
PODCAST BIBLE & BRAND CONTEXT:
|
||||
==============================
|
||||
{bible_context}
|
||||
|
||||
PODCAST ANALYSIS (from AI Analysis phase):
|
||||
==========================================
|
||||
{analysis_context}
|
||||
|
||||
RESEARCH DATA (from {len(sources)} sources):
|
||||
============================================
|
||||
{raw_content}
|
||||
|
||||
TASK:
|
||||
1. Provide a comprehensive summary (2-3 paragraphs) of the most important findings. Use Markdown for formatting (bolding, lists).
|
||||
2. Extract 3-5 "Key Insights". Each insight should have a title and a detailed explanation.
|
||||
3. For each insight, identify which source indices (e.g. 1, 2) it was derived from.
|
||||
YOUR TASK:
|
||||
==========
|
||||
As a podcast research expert, analyze this data and create content that will:
|
||||
1. Engage the specific target audience identified above
|
||||
2. Support the episode hook and key takeaways already planned
|
||||
3. Provide talking points that complement the guest's expertise
|
||||
4. Include a compelling call-to-action for listeners
|
||||
|
||||
NOTE: The research data includes "Key Highlights", "Summaries", and "Excerpts" from various sources.
|
||||
Pay special attention to the "Key Highlights" sections as they contain the most relevant information extracted by the neural search engine.
|
||||
|
||||
Return JSON structure:
|
||||
REQUIRED OUTPUT (JSON):
|
||||
=======================
|
||||
{{
|
||||
"summary": "Detailed markdown summary...",
|
||||
"summary": "2-3 paragraph comprehensive summary in Markdown. Start with a hook that matches the episode intro. Include specific data points, expert quotes, and trends.",
|
||||
"key_insights": [
|
||||
{{
|
||||
"title": "Insight Title",
|
||||
"content": "Detailed markdown content...",
|
||||
"source_indices": [1, 2]
|
||||
"title": "Catchy, engaging title for this insight",
|
||||
"content": "3-4 sentences with specific facts, quotes, or data. Write in a conversational tone suitable for a podcast host to discuss.",
|
||||
"source_indices": [1, 2, 3],
|
||||
"podcast_talking_points": ["Point 1 host can expand on", "Counter-point or follow-up", "Question to ask guest"]
|
||||
}}
|
||||
]
|
||||
],
|
||||
"expert_quotes": [
|
||||
{{
|
||||
"quote": "Direct quote from source",
|
||||
"source_index": 1,
|
||||
"context": "Why this quote matters for the podcast"
|
||||
}}
|
||||
],
|
||||
"listener_cta_suggestions": ["Specific action listener can take", "Resource to share", "Next episode preview"]
|
||||
}}
|
||||
|
||||
Requirements:
|
||||
- Ensure insights are deep, not just superficial facts. Look for trends, expert opinions, and specific data points.
|
||||
- Tone should be professional, insightful, and ready for a podcast host to discuss.
|
||||
- Avoid generic filler.
|
||||
QUALITY STANDARDS:
|
||||
==================
|
||||
- INSIGHTS MUST BE DEEP, not superficial - avoid generic statements
|
||||
- Include SPECIFIC DATA POINTS, percentages, statistics when available
|
||||
- Extract EXPERT QUOTES that hosts can reference
|
||||
- Identify GAPS in the research where more depth is needed
|
||||
- Make content naturally flow into the planned episode hook and CTA
|
||||
- Write in a CONVERSATIONAL tone - how a host would actually speak
|
||||
- Flag any CONTROVERSIAL or debatable claims for host to address
|
||||
"""
|
||||
try:
|
||||
logger.warning(f"[Podcast Research] Calling LLM for insight extraction...")
|
||||
llm_response = llm_text_gen(
|
||||
prompt=prompt,
|
||||
user_id=user_id,
|
||||
@@ -151,15 +199,45 @@ Requirements:
|
||||
preferred_provider=None,
|
||||
flow_type="premium_tool",
|
||||
)
|
||||
logger.warning(f"[Podcast Research] LLM response received, length: {len(llm_response) if llm_response else 0}")
|
||||
|
||||
# Normalize response
|
||||
# Normalize response - handle both string and dict responses
|
||||
data = None
|
||||
if isinstance(llm_response, str):
|
||||
data = json.loads(llm_response)
|
||||
try:
|
||||
# Try to fix common JSON issues
|
||||
fixed_response = llm_response.strip()
|
||||
# Remove markdown code blocks if present
|
||||
if fixed_response.startswith("```"):
|
||||
fixed_response = fixed_response.split("```")[1]
|
||||
if fixed_response.startswith("json"):
|
||||
fixed_response = fixed_response[4:]
|
||||
fixed_response = fixed_response.strip()
|
||||
data = json.loads(fixed_response)
|
||||
except json.JSONDecodeError as json_err:
|
||||
logger.warning(f"[Podcast Research] Failed to parse JSON: {json_err}. Response preview: {llm_response[:500]}...")
|
||||
# Try to extract JSON from response using regex
|
||||
json_match = re.search(r'\{.*\}', llm_response, re.DOTALL)
|
||||
if json_match:
|
||||
try:
|
||||
data = json.loads(json_match.group())
|
||||
logger.warning("[Podcast Research] Successfully extracted JSON via regex")
|
||||
except:
|
||||
pass
|
||||
else:
|
||||
data = llm_response
|
||||
|
||||
summary = data.get("summary", "")
|
||||
key_insights = [PodcastResearchInsight(**insight) for insight in data.get("key_insights", [])]
|
||||
|
||||
if data:
|
||||
try:
|
||||
summary = data.get("summary", "")
|
||||
key_insights = [PodcastResearchInsight(**insight) for insight in data.get("key_insights", [])]
|
||||
except Exception as insight_err:
|
||||
logger.warning(f"[Podcast Research] Failed to parse insights: {insight_err}. Data keys: {list(data.keys()) if isinstance(data, dict) else 'not a dict'}")
|
||||
summary = data.get("summary", "") if isinstance(data, dict) else ""
|
||||
key_insights = []
|
||||
else:
|
||||
summary = ""
|
||||
key_insights = []
|
||||
except HTTPException:
|
||||
raise
|
||||
except Exception as exc:
|
||||
@@ -183,21 +261,32 @@ Requirements:
|
||||
logger.warning(f"[Podcast Exa Research] Failed to track usage: {track_err}")
|
||||
|
||||
sources_payload = []
|
||||
seen_urls = set()
|
||||
for src in sources:
|
||||
url = src.get("url", "")
|
||||
# Skip duplicates
|
||||
if url and url in seen_urls:
|
||||
continue
|
||||
if url:
|
||||
seen_urls.add(url)
|
||||
|
||||
try:
|
||||
sources_payload.append(PodcastExaSource(**src))
|
||||
except Exception:
|
||||
sources_payload.append(PodcastExaSource(**{
|
||||
"title": src.get("title", ""),
|
||||
"url": src.get("url", ""),
|
||||
"excerpt": src.get("excerpt", ""),
|
||||
"url": url,
|
||||
"excerpt": src.get("excerpt") or (src.get("highlights")[0] if src.get("highlights") else "") or src.get("summary", ""),
|
||||
"published_at": src.get("published_at"),
|
||||
"publishedDate": src.get("publishedDate"),
|
||||
"highlights": src.get("highlights"),
|
||||
"summary": src.get("summary"),
|
||||
"source_type": src.get("source_type"),
|
||||
"index": src.get("index"),
|
||||
"image": src.get("image"),
|
||||
"author": src.get("author"),
|
||||
"text": src.get("text"),
|
||||
"credibility_score": src.get("credibility_score"),
|
||||
}))
|
||||
|
||||
return PodcastExaResearchResponse(
|
||||
|
||||
@@ -1,11 +1,12 @@
|
||||
"""
|
||||
Podcast Script Handlers
|
||||
|
||||
Script generation endpoint.
|
||||
Script generation and approval endpoints.
|
||||
"""
|
||||
|
||||
from fastapi import APIRouter, Depends, HTTPException
|
||||
from typing import Dict, Any
|
||||
from typing import Dict, Any, Optional
|
||||
from pydantic import BaseModel, Field
|
||||
import json
|
||||
|
||||
from middleware.auth_middleware import get_current_user
|
||||
@@ -24,6 +25,29 @@ from ..models import (
|
||||
router = APIRouter()
|
||||
|
||||
|
||||
class SceneApprovalRequest(BaseModel):
|
||||
project_id: str = Field(..., min_length=1)
|
||||
scene_id: str = Field(..., min_length=1)
|
||||
approved: bool = True
|
||||
notes: Optional[str] = None
|
||||
|
||||
|
||||
@router.post("/script/approve")
|
||||
async def approve_podcast_scene(
|
||||
request: SceneApprovalRequest,
|
||||
current_user: Dict[str, Any] = Depends(get_current_user),
|
||||
) -> Dict[str, Any]:
|
||||
"""Persist scene approval metadata for auditing (podcast-specific)."""
|
||||
user_id = require_authenticated_user(current_user)
|
||||
logger.warning(f"[Podcast] Scene approval recorded user={user_id} project={request.project_id} scene={request.scene_id} approved={request.approved}")
|
||||
return {
|
||||
"success": True,
|
||||
"project_id": request.project_id,
|
||||
"scene_id": request.scene_id,
|
||||
"approved": request.approved,
|
||||
}
|
||||
|
||||
|
||||
@router.post("/script", response_model=PodcastScriptResponse)
|
||||
async def generate_podcast_script(
|
||||
request: PodcastScriptRequest,
|
||||
@@ -33,6 +57,10 @@ async def generate_podcast_script(
|
||||
Generate a podcast script outline (scenes + lines) using podcast-oriented prompting.
|
||||
"""
|
||||
user_id = require_authenticated_user(current_user)
|
||||
logger.warning(f"[ScriptGen] ========== SCRIPT GENERATION START ==========")
|
||||
logger.warning(f"[ScriptGen] Topic: {request.idea[:60]}...")
|
||||
logger.warning(f"[ScriptGen] Duration: {request.duration_minutes} min, Speakers: {request.speakers}")
|
||||
logger.warning(f"[ScriptGen] Has research: {bool(request.research)}, Has bible: {bool(request.bible)}, Has analysis: {bool(request.analysis)}")
|
||||
|
||||
# Build comprehensive research context for higher-quality scripts
|
||||
research_context = ""
|
||||
@@ -77,55 +105,53 @@ async def generate_podcast_script(
|
||||
# Extract Analysis and Outline context for grounding
|
||||
analysis_context = ""
|
||||
if request.analysis:
|
||||
analysis_context = f"""
|
||||
TARGET AUDIENCE: {request.analysis.get('audience', 'General')}
|
||||
CONTENT TYPE: {request.analysis.get('contentType', 'Conversational')}
|
||||
TOP KEYWORDS: {', '.join(request.analysis.get('topKeywords', []))}
|
||||
"""
|
||||
try:
|
||||
audience = request.analysis.get('audience', '') or ''
|
||||
content_type = request.analysis.get('contentType', '') or ''
|
||||
keywords = request.analysis.get('topKeywords', []) or []
|
||||
analysis_context = f"ANALYSIS: Audience={audience} | Type={content_type} | Keywords={', '.join(keywords[:8])}"
|
||||
except:
|
||||
pass
|
||||
|
||||
outline_context = ""
|
||||
if request.outline:
|
||||
outline_context = f"""
|
||||
REFINED EPISODE OUTLINE (Follow this structure closely):
|
||||
Title: {request.outline.get('title', 'N/A')}
|
||||
Segments: {' | '.join(request.outline.get('segments', []))}
|
||||
"""
|
||||
try:
|
||||
title = request.outline.get('title', '') or ''
|
||||
segments = request.outline.get('segments', []) or []
|
||||
outline_context = f"OUTLINE: {title} - {' | '.join(segments[:5])}"
|
||||
except:
|
||||
pass
|
||||
|
||||
prompt = f"""You are an expert podcast script planner. Create natural, conversational podcast scenes.
|
||||
prompt = f"""Create a podcast script with scenes and dialogue.
|
||||
|
||||
{f"PODCAST BIBLE (Hyper-Personalization Context):\n{bible_context}\n" if bible_context else ""}
|
||||
{f"ANALYSIS CONTEXT:\n{analysis_context}\n" if analysis_context else ""}
|
||||
{f"REFINED OUTLINE:\n{outline_context}\n" if outline_context else ""}
|
||||
{f"BIBLE: {bible_context[:1500]}" if bible_context else ""}
|
||||
{f"{analysis_context}" if analysis_context else ""}
|
||||
{f"{outline_context}" if outline_context else ""}
|
||||
{f"RESEARCH: {research_context[:1200]}" if research_context else ""}
|
||||
|
||||
Podcast Idea: "{request.idea}"
|
||||
Duration: ~{request.duration_minutes} minutes
|
||||
Speakers: {request.speakers} (Host + optional Guest)
|
||||
Topic: "{request.idea}"
|
||||
Duration: {request.duration_minutes} min | Speakers: {request.speakers}
|
||||
|
||||
{f"RESEARCH CONTEXT:\n{research_context}\n" if research_context else ""}
|
||||
Return JSON with scenes array. Each scene:
|
||||
- id: string
|
||||
- title: short title (<=50 chars)
|
||||
- duration: seconds (total/5)
|
||||
- emotion: neutral|happy|excited|serious|curious|confident
|
||||
- lines: array of {{speaker, text, emphasis}}
|
||||
- Use 2-4 LINES PER SCENE (shorter script = lower TTS costs)
|
||||
- Each line: 1-3 sentences, conversational
|
||||
- Plain text only, no markdown
|
||||
|
||||
Return JSON with:
|
||||
- scenes: array of scenes. Each scene has:
|
||||
- id: string
|
||||
- title: short scene title (<= 60 chars)
|
||||
- duration: duration in seconds (evenly split across total duration)
|
||||
- emotion: string (one of: "neutral", "happy", "excited", "serious", "curious", "confident")
|
||||
- lines: array of {{"speaker": "...", "text": "...", "emphasis": boolean}}
|
||||
* Write natural, conversational dialogue
|
||||
* Each line can be a sentence or a few sentences that flow together
|
||||
* Use plain text only - no markdown formatting (no asterisks, underscores, etc.)
|
||||
* Mark "emphasis": true for key statistics or important points
|
||||
|
||||
Guidelines:
|
||||
- Write for spoken delivery: conversational, natural, with contractions.
|
||||
- Follow the interaction tone specified in the Bible.
|
||||
- Ensure the Host persona matches the background and personality traits from the Bible.
|
||||
- Structure the intro and outro scenes according to the Bible's "Intro Format" and "Outro Format".
|
||||
- Adhere to any constraints mentioned in the Bible.
|
||||
- Use insights from the Research Context to ground the conversation in facts.
|
||||
- IMPORTANT: Follow the REFINED OUTLINE segments as the primary structure for the episode.
|
||||
COST OPTIMIZATION:
|
||||
- 5-6 scenes max for {request.duration_minutes} min episode
|
||||
- Concise, information-dense dialogue
|
||||
- Skip filler words and redundant phrases
|
||||
- Focus on unique insights from research
|
||||
- Make every line count toward value delivery
|
||||
"""
|
||||
|
||||
try:
|
||||
logger.warning(f"[ScriptGen] Calling LLM to generate script (prompt length: {len(prompt)})...")
|
||||
raw = llm_text_gen(
|
||||
prompt=prompt,
|
||||
user_id=user_id,
|
||||
@@ -133,6 +159,7 @@ Guidelines:
|
||||
preferred_provider=None,
|
||||
flow_type="premium_tool",
|
||||
)
|
||||
logger.warning(f"[ScriptGen] LLM response received, length: {len(raw) if raw else 0}")
|
||||
except HTTPException:
|
||||
raise
|
||||
except Exception as exc:
|
||||
|
||||
@@ -140,17 +140,20 @@ def _execute_podcast_video_task(
|
||||
except Exception as e:
|
||||
logger.warning(f"[Podcast] Failed to fetch project context for video generation: {e}")
|
||||
|
||||
# Prepare scene data for animation
|
||||
# Prepare scene data for animation - include all context for enhanced prompt
|
||||
scene_data = {
|
||||
"scene_number": scene_number,
|
||||
"title": request.scene_title,
|
||||
"scene_id": request.scene_id,
|
||||
"image_prompt": request.scene_image_prompt,
|
||||
"description": request.scene_narration,
|
||||
"lines": [{"text": request.scene_narration}] if request.scene_narration else [],
|
||||
}
|
||||
story_context = {
|
||||
"project_id": request.project_id,
|
||||
"type": "podcast",
|
||||
"bible": project_bible,
|
||||
"analysis": project_analysis,
|
||||
"analysis": request.analysis or project_analysis, # Use passed analysis or fallback to DB
|
||||
}
|
||||
|
||||
animation_result = animate_scene_with_voiceover(
|
||||
|
||||
@@ -63,6 +63,10 @@ class PodcastAnalyzeResponse(BaseModel):
|
||||
top_keywords: list[str]
|
||||
suggested_outlines: list[Dict[str, Any]]
|
||||
title_suggestions: list[str]
|
||||
episode_hook: Optional[str] = None
|
||||
key_takeaways: Optional[list[str]] = None
|
||||
guest_talking_points: Optional[list[str]] = None
|
||||
listener_cta: Optional[str] = None
|
||||
research_queries: Optional[List[Dict[str, str]]] = None
|
||||
exa_suggested_config: Optional[Dict[str, Any]] = None
|
||||
bible: Optional[Dict[str, Any]] = None
|
||||
@@ -142,12 +146,15 @@ class PodcastExaSource(BaseModel):
|
||||
url: str = ""
|
||||
excerpt: str = ""
|
||||
published_at: Optional[str] = None
|
||||
publishedDate: Optional[str] = None # Exa format
|
||||
highlights: Optional[List[str]] = None
|
||||
summary: Optional[str] = None
|
||||
source_type: Optional[str] = None
|
||||
index: Optional[int] = None
|
||||
image: Optional[str] = None
|
||||
author: Optional[str] = None
|
||||
text: Optional[str] = None # Exa full text
|
||||
credibility_score: Optional[float] = None # Exa scores
|
||||
|
||||
|
||||
class PodcastResearchInsight(BaseModel):
|
||||
@@ -155,6 +162,9 @@ class PodcastResearchInsight(BaseModel):
|
||||
title: str
|
||||
content: str
|
||||
source_indices: List[int] = []
|
||||
podcast_talking_points: Optional[List[str]] = [] # Talking points for host to expand on
|
||||
expert_quotes: Optional[List[Dict[str, str]]] = [] # Quotes from sources
|
||||
listener_cta_suggestions: Optional[List[str]] = [] # CTA suggestions
|
||||
|
||||
|
||||
class PodcastExaResearchResponse(BaseModel):
|
||||
@@ -178,6 +188,7 @@ class PodcastAudioRequest(BaseModel):
|
||||
scene_title: str
|
||||
text: str
|
||||
voice_id: Optional[str] = "Wise_Woman"
|
||||
custom_voice_id: Optional[str] = None # Voice clone ID for custom voice
|
||||
speed: Optional[float] = 1.0
|
||||
volume: Optional[float] = 1.0
|
||||
pitch: Optional[float] = 0.0
|
||||
@@ -263,7 +274,9 @@ class PodcastImageRequest(BaseModel):
|
||||
scene_id: str
|
||||
scene_title: str
|
||||
scene_content: Optional[str] = None # Optional: scene lines text for context
|
||||
scene_emotion: Optional[str] = None # Optional: scene emotion for visual tone
|
||||
idea: Optional[str] = None # Optional: podcast idea for context
|
||||
analysis: Optional[Dict[str, Any]] = Field(None, description="AI analysis for visual context (keywords, audience)")
|
||||
base_avatar_url: Optional[str] = None # Base avatar image URL for scene variations
|
||||
bible: Optional[Dict[str, Any]] = Field(None, description="Podcast Bible for hyper-personalization")
|
||||
width: int = 1024
|
||||
@@ -285,6 +298,7 @@ class PodcastImageResponse(BaseModel):
|
||||
provider: str
|
||||
model: Optional[str] = None
|
||||
cost: float
|
||||
image_prompt: Optional[str] = None # Return the prompt used for generation
|
||||
|
||||
|
||||
class PodcastVideoGenerationRequest(BaseModel):
|
||||
@@ -295,6 +309,9 @@ class PodcastVideoGenerationRequest(BaseModel):
|
||||
audio_url: str = Field(..., description="URL to the generated audio file")
|
||||
avatar_image_url: Optional[str] = Field(None, description="URL to scene image (required for video generation)")
|
||||
bible: Optional[Dict[str, Any]] = Field(None, description="Podcast Bible for hyper-personalization")
|
||||
analysis: Optional[Dict[str, Any]] = Field(None, description="Podcast Analysis for context (content type, audience, takeaways, guest)")
|
||||
scene_image_prompt: Optional[str] = Field(None, description="Original image generation prompt for visual context")
|
||||
scene_narration: Optional[str] = Field(None, description="Scene narration/script lines for context")
|
||||
resolution: str = Field("720p", description="Video resolution (480p or 720p)")
|
||||
prompt: Optional[str] = Field(None, description="Optional animation prompt override")
|
||||
seed: Optional[int] = Field(-1, description="Random seed; -1 for random")
|
||||
|
||||
Reference in New Issue
Block a user