update

app.py

@@ -194,33 +194,73 @@ def search_pexels_video_tool(style: str, niche: str) -> dict:
         }
 
 @tool
-def generate_voice_narration_tool(quote_text: str, output_path: str) -> dict:
+def generate_voice_commentary_tool(quote_text: str, niche: str, output_path: str) -> dict:
     """
-    Generate voice narration for the quote using ElevenLabs.
+    Generate insightful voice commentary explaining the deeper meaning of the quote.
+    Uses Gemini to create thoughtful explanation, then ElevenLabs to voice it.
+    This adds VALUE - not just reading what's already on screen.
     
     Args:
-        quote_text: The quote text to narrate
+        quote_text: The quote to explain
+        niche: The niche/category for context
         output_path: Path where to save the audio file
     
     Returns:
-        Dictionary with success status and output path
+        Dictionary with success status, output path, and the explanation text
     """
     
     try:
-        # Generate audio using ElevenLabs
+        # Step 1: Generate explanation using Gemini
+        import google.generativeai as genai
+        
+        explanation_prompt = f"""Given this {niche} quote:
+
+"{quote_text}"
+
+Write a brief, insightful voice-over commentary that explains the deeper meaning or practical wisdom.
+
+Requirements:
+- 2-3 sentences maximum
+- Around 25-35 words total
+- Spoken naturally (like a wise mentor)
+- Add insight that isn't obvious from reading
+- Make it thought-provoking
+- Don't start with "This quote..." - dive into the insight
+
+Example:
+Quote: "Between stimulus and response there is a space."
+Good: "In that pause lies your freedom. That's where you choose who you become, not who your habits make you."
+
+Return ONLY the commentary, nothing else."""
+
+        genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
+        model = genai.GenerativeModel('gemini-pro')
+        
+        response = model.generate_content(
+            explanation_prompt,
+            generation_config={
+                "temperature": 0.7,
+                "max_output_tokens": 100
+            }
+        )
+        
+        explanation = response.text.strip().strip('"').strip("'")
+        print(f"📝 Commentary: {explanation}")
+        
+        # Step 2: Generate voice using ElevenLabs
         audio = elevenlabs_client.text_to_speech.convert(
-            text=quote_text,
-            voice_id="pNInz6obpgDQGcFmaJgB",  # Adam voice - clear and motivational
+            text=explanation,
+            voice_id="pNInz6obpgDQGcFmaJgB",  # Adam - thoughtful and clear
             model_id="eleven_multilingual_v2",
             voice_settings=VoiceSettings(
-                stability=0.5,
-                similarity_boost=0.75,
-                style=0.5,
+                stability=0.6,  # Stable for narration
+                similarity_boost=0.8,
+                style=0.6,  # Expressive for commentary
                 use_speaker_boost=True
             )
         )
         
-        # Save audio to file
+        # Save audio
         with open(output_path, 'wb') as f:
             for chunk in audio:
                 f.write(chunk)
@@ -228,14 +268,16 @@ def generate_voice_narration_tool(quote_text: str, output_path: str) -> dict:
         return {
             "success": True,
             "output_path": output_path,
-            "message": "Voice narration created successfully!"
+            "explanation": explanation,
+            "message": "Voice commentary created!"
         }
     
     except Exception as e:
         return {
             "success": False,
             "output_path": None,
-            "message": f"Error creating voice: {str(e)}"
+            "explanation": None,
+            "message": f"Error creating commentary: {str(e)}"
         }
 
 @tool
@@ -265,9 +307,13 @@ def create_quote_video_tool(video_url: str, quote_text: str, output_path: str, a
             
             print("🚀 Processing on Modal (fast!)...")
             
-            # For now, skip audio in Modal (would need to upload to cloud storage)
-            # We'll process without audio for speed
-            audio_url = None
+            # Prepare audio data if present
+            audio_b64 = None
+            if audio_path and os.path.exists(audio_path):
+                with open(audio_path, 'rb') as f:
+                    audio_bytes = f.read()
+                    audio_b64 = base64.b64encode(audio_bytes).decode()
+                print(f"   🎤 Including voice commentary audio ({len(audio_bytes)} bytes)")
             
             # Call Modal endpoint
             response = requests.post(
@@ -275,9 +321,9 @@ def create_quote_video_tool(video_url: str, quote_text: str, output_path: str, a
                 json={
                     "video_url": video_url,
                     "quote_text": quote_text,
-                    "audio_url": audio_url
+                    "audio_b64": audio_b64  # Pass audio as base64
                 },
-                timeout=300  # Increased to 5 minutes
+                timeout=300
             )
             
             if response.status_code == 200:
@@ -459,9 +505,9 @@ def initialize_agent():
         
         # Create agent with custom tools
         agent = CodeAgent(
-            tools=[generate_quote_tool, search_pexels_video_tool, generate_voice_narration_tool, create_quote_video_tool],
+            tools=[generate_quote_tool, search_pexels_video_tool, generate_voice_commentary_tool, create_quote_video_tool],
             model=model,
-            additional_authorized_imports=["requests", "openai", "random", "tempfile", "os"],
+            additional_authorized_imports=["requests", "openai", "random", "tempfile", "os", "google.generativeai"],
             max_steps=15
         )
         
@@ -476,16 +522,29 @@ def initialize_agent():
 # Initialize agent
 agent, agent_error = initialize_agent()
 
-def mcp_agent_pipeline(niche, style, num_variations=1):
+def mcp_agent_pipeline(niche, style, num_variations=1, add_voice=False):
     """
     MCP-POWERED AUTONOMOUS AGENT PIPELINE
     Uses smolagents with proper MCP server integration
     Generates multiple video variations with Gemini-powered quotes
+    Optionally adds ElevenLabs voice narration
     """
     
+    # Smart auto-enable voice for contemplative content
+    auto_voice_niches = ["Stoicism", "Mindfulness", "Leadership"]
+    if niche in auto_voice_niches and not add_voice:
+        add_voice = True
+        voice_reason = "🎤 Voice auto-enabled: Perfect for contemplative content!"
+    else:
+        voice_reason = None
+    
     status_log = []
     status_log.append("🤖 **MCP AGENT STARTING**\n")
     
+    if voice_reason:
+        status_log.append(voice_reason)
+        status_log.append("")
+    
     if agent_error:
         status_log.append(f"❌ Agent initialization failed: {agent_error}")
         status_log.append("\n🔄 Falling back to direct tool execution...\n")
@@ -523,9 +582,35 @@ def mcp_agent_pipeline(niche, style, num_variations=1):
         
         status_log.append("")
         
-        # STEP 4: Create multiple video variations
+        # STEP 4: Generate voice commentary if enabled
+        audio_path = None
+        explanation_text = None
+        if add_voice:
+            status_log.append("🎤 **GEMINI + ELEVENLABS: generate_voice_commentary_tool**")
+            status_log.append("   ⏳ Creating insightful commentary...")
+            
+            import tempfile
+            audio_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
+            audio_path = audio_file.name
+            
+            voice_result = generate_voice_commentary_tool(quote, niche, audio_path)
+            
+            if voice_result["success"]:
+                explanation_text = voice_result.get("explanation")
+                status_log.append("   ✅ Commentary created!")
+                status_log.append(f"   💭 Insight: \"{explanation_text}\"")
+                status_log.append(f"   🔊 Using Gemini (explanation) + ElevenLabs (voice)")
+            else:
+                status_log.append(f"   ⚠️ Commentary failed: {voice_result.get('message', 'Unknown error')}")
+                audio_path = None
+            
+            status_log.append("")
+        
+        # STEP 5: Create multiple video variations
         status_log.append(f"🎬 **MCP TOOL: create_quote_video_tool (x{len(video_results)})**")
         status_log.append(f"   ⏳ Creating {len(video_results)} video variations...")
+        if add_voice and audio_path:
+            status_log.append("   🎤 Including voice narration...")
         
         output_dir = "/tmp/quote_videos"
         gallery_dir = "/data/gallery_videos"  # HF persistent storage
@@ -544,7 +629,7 @@ def mcp_agent_pipeline(niche, style, num_variations=1):
                 video_result["video_url"],
                 quote,
                 output_path,
-                None  # No audio
+                audio_path  # Use voice if enabled
             )
             
             if creation_result["success"]:
@@ -569,10 +654,12 @@ def mcp_agent_pipeline(niche, style, num_variations=1):
         
         status_log.append("")
         
-        # STEP 5: Integration status
+        # STEP 6: Integration status
         status_log.append("🔗 **AI INTEGRATIONS:**")
         status_log.append("   ✅ Gemini API - Quote generation with variety tracking")
         status_log.append("   ✅ Pexels API - Video search")
+        if add_voice:
+            status_log.append("   ✅ ElevenLabs - Premium AI voice synthesis")
         status_log.append("   ✅ Modal Compute - Fast video processing")
         if mcp_enabled:
             status_log.append("   ✅ MCP Server - abidlabs-mcp-tools.hf.space")
@@ -590,7 +677,7 @@ def mcp_agent_pipeline(niche, style, num_variations=1):
         status_log.append(f"\n❌ Pipeline error: {str(e)}")
         return "\n".join(status_log), []
 
-def fallback_pipeline(niche, style, num_variations=1):
+def fallback_pipeline(niche, style, num_variations=1, add_voice=False):
     """Fallback pipeline if MCP agent fails"""
     status_log = []
     status_log.append("🔄 **FALLBACK MODE (Direct Tool Execution)**\n")
@@ -604,6 +691,20 @@ def fallback_pipeline(niche, style, num_variations=1):
     
     status_log.append(f"   ✅ Quote generated\n")
     
+    # Generate voice commentary if enabled
+    audio_path = None
+    if add_voice:
+        status_log.append("🎤 Generating commentary...")
+        import tempfile
+        audio_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
+        audio_path = audio_file.name
+        voice_result = generate_voice_commentary_tool(quote, niche, audio_path)
+        if voice_result["success"]:
+            status_log.append(f"   ✅ Commentary: {voice_result.get('explanation')}\n")
+        else:
+            audio_path = None
+            status_log.append("   ⚠️ Commentary failed\n")
+    
     # Search videos
     status_log.append(f"🔍 Searching for {num_variations} videos...")
     video_results = []
@@ -636,7 +737,7 @@ def fallback_pipeline(niche, style, num_variations=1):
             video_result["video_url"],
             quote,
             output_path,
-            None  # No audio
+            audio_path  # Use voice if enabled
         )
         
         if creation_result["success"]:
@@ -666,21 +767,23 @@ def fallback_pipeline(niche, style, num_variations=1):
 with gr.Blocks(title="AIQuoteClipGenerator - MCP + Gemini Edition", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # 🎬 AIQuoteClipGenerator
-    ### MCP-Powered with Gemini AI Integration
+    ### MCP-Powered with Gemini AI + ElevenLabs Voice Commentary
     
     **Key Features:**
     - 🌟 **Gemini AI:** No more repetitive quotes! Smart variety tracking
+    - 🎤 **Voice Commentary:** AI explains the deeper meaning (not just reading the quote!)
+    - 🧠 **Dual Gemini Use:** Quote generation + Explanation generation
+    - 🔊 **ElevenLabs Voice:** Premium AI voice synthesis
     - 🔗 **MCP Server:** smolagents framework integration  
-    - 🛠️ **4 Custom MCP Tools:** Quote (Gemini) + Video search + Voice + Video creation
-    - 🤖 **Agent Reasoning:** Autonomous task execution
+    - 🛠️ **4 Custom MCP Tools:** Quote + Video search + Commentary + Video creation
     - ⚡ **Modal Processing:** 4-8x faster video creation
     - 🎨 **Multiple Variations:** Get different video styles
     
     **Prize Eligibility:**
-    - ✅ Gemini API Integration ($10K Creative category)
-    - ✅ OpenAI Fallback (API Integration $1K credits)
+    - ✅ Gemini API Integration ($10K Creative category) - Used TWICE!
+    - ✅ ElevenLabs Voice Award (~$2K + AirPods 4 Pro) - Adds real value!
     - ✅ Modal Innovation Award ($2.5K)
-    - ✅ ElevenLabs Voice Award (~$2K + AirPods)
+    - ✅ OpenAI Fallback ($1K credits)
     """)
     
     # Example Gallery - Instagram-style grid
@@ -757,6 +860,12 @@ with gr.Blocks(title="AIQuoteClipGenerator - MCP + Gemini Edition", theme=gr.the
                 info="Generate multiple versions to choose from"
             )
             
+            add_voice = gr.Checkbox(
+                label="🎤 Add Voice Commentary (Gemini + ElevenLabs)",
+                value=False,
+                info="AI explains the quote's deeper meaning with voice - adds real value! Auto-enabled for Stoicism/Mindfulness."
+            )
+            
             generate_btn = gr.Button("🤖 Run MCP Agent with Gemini", variant="primary", size="lg")
         
         with gr.Column():
@@ -784,15 +893,10 @@ with gr.Blocks(title="AIQuoteClipGenerator - MCP + Gemini Edition", theme=gr.the
     **Category:** Productivity Tools  
     **Built with:** Gradio + smolagents + Gemini + OpenAI + Pexels + Modal + ElevenLabs + MCP
     
-    **Prize Targets:**
-    - Google Gemini Creative Award ($10K)
-    - Modal Innovation Award ($2.5K)
-    - OpenAI API Integration ($1K credits)
-    - ElevenLabs Voice Award (~$2K + AirPods)
     """)
     
-    def process_and_display(niche, style, num_variations):
-        status, videos = mcp_agent_pipeline(niche, style, num_variations)
+    def process_and_display(niche, style, num_variations, add_voice):
+        status, videos = mcp_agent_pipeline(niche, style, num_variations, add_voice)
         
         # Return up to 3 videos, None for unused slots
         v1 = videos[0] if len(videos) > 0 else None
@@ -806,7 +910,7 @@ with gr.Blocks(title="AIQuoteClipGenerator - MCP + Gemini Edition", theme=gr.the
     
     generate_btn.click(
         process_and_display, 
-        inputs=[niche, style, num_variations], 
+        inputs=[niche, style, num_variations, add_voice], 
         outputs=[
             output, video1, video2, video3,
             gallery_video1, gallery_video2, gallery_video3,

modal_video_processing.py

@@ -26,7 +26,7 @@ image = modal.Image.debian_slim(python_version="3.11").pip_install(
     keep_warm=1,  # Keep 1 container warm to eliminate cold starts!
     container_idle_timeout=300,  # Keep alive for 5 minutes
 )
-def process_quote_video(video_url: str, quote_text: str, audio_url: str = None) -> bytes:
+def process_quote_video(video_url: str, quote_text: str, audio_b64: str = None) -> bytes:
     """
     Process quote video on Modal's fast infrastructure.
     Downloads video, adds text overlay, optionally adds audio, returns video bytes.
@@ -34,7 +34,7 @@ def process_quote_video(video_url: str, quote_text: str, audio_url: str = None)
     Args:
         video_url: URL of background video
         quote_text: Quote to overlay
-        audio_url: Optional URL of audio file
+        audio_b64: Optional base64-encoded audio file
     
     Returns:
         bytes: Processed video file as bytes
@@ -45,6 +45,7 @@ def process_quote_video(video_url: str, quote_text: str, audio_url: str = None)
     from PIL import Image, ImageDraw, ImageFont
     import numpy as np
     import time
+    import base64
     
     start_time = time.time()
     print(f"🎬 Starting video processing on Modal...")
@@ -146,11 +147,37 @@ def process_quote_video(video_url: str, quote_text: str, audio_url: str = None)
     final_video = CompositeVideoClip([video, text_clip])
     print(f"✅ Composited in {time.time() - composite_start:.1f}s")
     
-    # Skip audio for speed
-    if audio_url:
-        print("⚠️ Skipping audio for speed optimization")
+    # Add audio if provided
+    if audio_b64:
+        print("🎤 Adding voice commentary audio...")
+        audio_start = time.time()
+        try:
+            # Decode base64 audio
+            audio_bytes = base64.b64decode(audio_b64)
+            
+            # Save to temp file
+            temp_audio = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
+            with open(temp_audio.name, 'wb') as f:
+                f.write(audio_bytes)
+            
+            # Load audio clip
+            audio_clip = AudioFileClip(temp_audio.name)
+            
+            # Use the shorter duration between video and audio
+            audio_duration = min(audio_clip.duration, final_video.duration)
+            audio_clip = audio_clip.subclip(0, audio_duration)
+            
+            # Set audio on video
+            final_video = final_video.set_audio(audio_clip)
+            
+            print(f"✅ Audio added in {time.time() - audio_start:.1f}s")
+            
+            # Cleanup audio temp file
+            os.unlink(temp_audio.name)
+        except Exception as e:
+            print(f"⚠️ Audio failed: {e}, continuing without audio")
     
-    # Export with FASTEST possible settings
+    # Export with optimized settings
     print("📦 Exporting video...")
     export_start = time.time()
     output_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp4')
@@ -159,13 +186,13 @@ def process_quote_video(video_url: str, quote_text: str, audio_url: str = None)
         output_file.name,
         codec='libx264',
         audio_codec='aac',
-        fps=10,  # Even lower - 10fps is fine for quote videos
+        fps=10,  # Lower fps for speed
         preset='ultrafast',
         threads=2,
         verbose=False,
         logger=None,
-        bitrate="400k",  # Lower bitrate
-        ffmpeg_params=['-crf', '30', '-g', '30']  # Even lower quality, larger GOP
+        bitrate="400k",
+        ffmpeg_params=['-crf', '30', '-g', '30']
     )
     
     print(f"✅ Video exported in {time.time() - export_start:.1f}s")
@@ -192,18 +219,18 @@ def process_quote_video(video_url: str, quote_text: str, audio_url: str = None)
 @modal.web_endpoint(method="POST")
 def process_video_endpoint(data: dict):
     """
-    Web endpoint to process videos.
-    Accepts JSON with video_url, quote_text, and optional audio_url.
+    Web endpoint to process videos with optional audio.
+    Accepts JSON with video_url, quote_text, and optional audio_b64.
     """
     video_url = data.get("video_url")
     quote_text = data.get("quote_text")
-    audio_url = data.get("audio_url")
+    audio_b64 = data.get("audio_b64")  # Changed from audio_url
     
     if not video_url or not quote_text:
         return {"error": "Missing video_url or quote_text"}, 400
     
     try:
-        video_bytes = process_quote_video.remote(video_url, quote_text, audio_url)
+        video_bytes = process_quote_video.remote(video_url, quote_text, audio_b64)
         
         # Return video bytes as base64
         import base64
@@ -225,6 +252,6 @@ if __name__ == "__main__":
         result = process_quote_video.remote(
             video_url="https://videos.pexels.com/video-files/3843433/3843433-uhd_2732_1440_25fps.mp4",
             quote_text="Test quote for local testing",
-            audio_url=None
+            audio_b64=None
         )
         print(f"Got video: {len(result)} bytes")

quote_generator_gemini.py

@@ -283,4 +283,5 @@ if __name__ == "__main__":
         
     except Exception as e:
         print(f"Error: {e}")
-        print("\nMake sure GEMINI_API_KEY is set in environment variables")
+        print("\nMake sure GEMINI_API_KEY is set in environment variables")
+