Add ElevenLabs voice + multiple variations

app.py

@@ -6,14 +6,16 @@ import tempfile
 from openai import OpenAI
 from smolagents import CodeAgent, MCPClient, tool
 from huggingface_hub import InferenceClient
-from moviepy.editor import VideoFileClip, ImageClip, CompositeVideoClip
+from moviepy.editor import VideoFileClip, ImageClip, CompositeVideoClip, AudioFileClip
 from PIL import Image, ImageDraw, ImageFont
 import textwrap
 import numpy as np
+from elevenlabs import ElevenLabs, VoiceSettings
 
 # Initialize clients
 openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
 PEXELS_API_KEY = os.getenv("PEXELS_API_KEY")
+elevenlabs_client = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))
 
 # Initialize MCP Client (connecting to existing MCP server)
 try:
@@ -190,15 +192,62 @@ def search_pexels_video_tool(style: str, niche: str) -> dict:
         }
 
 @tool
-def create_quote_video_tool(video_url: str, quote_text: str, output_path: str) -> dict:
+def generate_voice_narration_tool(quote_text: str, output_path: str) -> dict:
+    """
+    Generate voice narration for the quote using ElevenLabs.
+    
+    Args:
+        quote_text: The quote text to narrate
+        output_path: Path where to save the audio file
+    
+    Returns:
+        Dictionary with success status and output path
+    """
+    
+    try:
+        # Generate audio using ElevenLabs
+        audio = elevenlabs_client.text_to_speech.convert(
+            text=quote_text,
+            voice_id="pNInz6obpgDQGcFmaJgB",  # Adam voice - clear and motivational
+            model_id="eleven_multilingual_v2",
+            voice_settings=VoiceSettings(
+                stability=0.5,
+                similarity_boost=0.75,
+                style=0.5,
+                use_speaker_boost=True
+            )
+        )
+        
+        # Save audio to file
+        with open(output_path, 'wb') as f:
+            for chunk in audio:
+                f.write(chunk)
+        
+        return {
+            "success": True,
+            "output_path": output_path,
+            "message": "Voice narration created successfully!"
+        }
+    
+    except Exception as e:
+        return {
+            "success": False,
+            "output_path": None,
+            "message": f"Error creating voice: {str(e)}"
+        }
+
+@tool
+def create_quote_video_tool(video_url: str, quote_text: str, output_path: str, audio_path: str = None) -> dict:
     """
     Create a final quote video by overlaying text on the background video.
     Uses PIL/Pillow for text rendering (works on Hugging Face Spaces).
+    Optionally adds voice narration audio.
     
     Args:
         video_url: URL of the background video from Pexels
         quote_text: The quote text to overlay
         output_path: Path where to save the final video
+        audio_path: Optional path to audio file for voice narration
     
     Returns:
         Dictionary with success status and output path
@@ -304,6 +353,17 @@ def create_quote_video_tool(video_url: str, quote_text: str, output_path: str) -
         # Step 5: Composite video with text
         final_video = CompositeVideoClip([video, text_clip])
         
+        # Step 5.5: Add voice narration if provided
+        if audio_path and os.path.exists(audio_path):
+            try:
+                audio_clip = AudioFileClip(audio_path)
+                # 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)
+                final_video = final_video.set_audio(audio_clip)
+            except Exception as audio_error:
+                print(f"Warning: Could not add audio: {audio_error}")
+        
         # Step 6: Export final video
         final_video.write_videofile(
             output_path,
@@ -343,10 +403,10 @@ def initialize_agent():
         
         # Create agent with custom tools
         agent = CodeAgent(
-            tools=[generate_quote_tool, search_pexels_video_tool, create_quote_video_tool],
+            tools=[generate_quote_tool, search_pexels_video_tool, generate_voice_narration_tool, create_quote_video_tool],
             model=model,
             additional_authorized_imports=["requests", "openai", "random", "tempfile", "os"],
-            max_steps=10
+            max_steps=15
         )
         
         # Add MCP client if available
@@ -360,10 +420,11 @@ def initialize_agent():
 # Initialize agent
 agent, agent_error = initialize_agent()
 
-def mcp_agent_pipeline(niche, style):
+def mcp_agent_pipeline(niche, style, num_variations=3, add_voice=True):
     """
     MCP-POWERED AUTONOMOUS AGENT PIPELINE
     Uses smolagents with proper MCP server integration
+    Generates multiple video variations with optional voice narration
     """
     
     status_log = []
@@ -372,85 +433,120 @@ def mcp_agent_pipeline(niche, style):
     if agent_error:
         status_log.append(f"❌ Agent initialization failed: {agent_error}")
         status_log.append("\n🔄 Falling back to direct tool execution...\n")
-        return fallback_pipeline(niche, style)
+        return fallback_pipeline(niche, style, num_variations, add_voice)
     
     try:
         # STEP 1: Agent receives task
         status_log.append("📋 **TASK RECEIVED:**")
         status_log.append(f"   → Generate {niche} quote with {style} aesthetic")
-        status_log.append(f"   → Find matching video")
-        status_log.append(f"   → Create final quote video\n")
+        status_log.append(f"   → Create {num_variations} video variations")
+        if add_voice:
+            status_log.append(f"   → Add voice narration with ElevenLabs")
+        status_log.append("")
         
         # STEP 2: Agent executes quote generation
         status_log.append("🧠 **MCP TOOL: generate_quote_tool**")
         quote = generate_quote_tool(niche, style)
         
         if "Error" in quote:
-            return "\n".join(status_log) + f"\n❌ Failed: {quote}", None, None
+            return "\n".join(status_log) + f"\n❌ Failed: {quote}", None, []
         
-        status_log.append(f"   ✅ Generated: \"{quote}\"\n")
-        
-        # STEP 3: Agent executes video search
-        status_log.append("🔍 **MCP TOOL: search_pexels_video_tool**")
-        video_result = search_pexels_video_tool(style, niche)
+        status_log.append(f"   ✅ Generated: \"{quote[:100]}...\"" if len(quote) > 100 else f"   ✅ Generated: \"{quote}\"\n")
         
-        if not video_result["success"]:
-            error_msg = video_result.get("error", "Unknown error")
-            return "\n".join(status_log) + f"\n❌ Video search failed: {error_msg}", None, None
+        # STEP 3: Generate voice narration if requested
+        audio_path = None
+        if add_voice:
+            status_log.append("🎤 **MCP TOOL: generate_voice_narration_tool**")
+            status_log.append("   ⏳ Creating AI voice narration...")
+            
+            audio_dir = "/tmp/quote_audio"
+            os.makedirs(audio_dir, exist_ok=True)
+            
+            import time
+            audio_filename = f"narration_{int(time.time())}.mp3"
+            audio_path = os.path.join(audio_dir, audio_filename)
+            
+            voice_result = generate_voice_narration_tool(quote, audio_path)
+            
+            if voice_result["success"]:
+                status_log.append(f"   ✅ Voice narration created!\n")
+            else:
+                status_log.append(f"   ⚠️ Voice creation failed, continuing without audio\n")
+                audio_path = None
+        
+        # STEP 4: Search for multiple videos
+        status_log.append(f"🔍 **MCP TOOL: search_pexels_video_tool (x{num_variations})**")
+        status_log.append(f"   ⏳ Finding {num_variations} different videos...")
+        
+        video_results = []
+        for i in range(num_variations):
+            video_result = search_pexels_video_tool(style, niche)
+            if video_result["success"]:
+                video_results.append(video_result)
+                status_log.append(f"   ✅ Video {i+1}: {video_result['search_query']}")
+        
+        if not video_results:
+            return "\n".join(status_log) + "\n❌ No videos found", None, []
         
-        status_log.append(f"   ✅ Found video: {video_result['search_query']}")
-        status_log.append(f"   📥 Video URL: {video_result['video_url']}\n")
+        status_log.append("")
         
-        # STEP 4: Agent creates final video with text overlay
-        status_log.append("🎬 **MCP TOOL: create_quote_video_tool**")
-        status_log.append("   ⏳ Creating video with text overlay...")
+        # 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...")
         
-        # Create output directory if it doesn't exist
         output_dir = "/tmp/quote_videos"
         os.makedirs(output_dir, exist_ok=True)
         
-        # Generate unique filename
+        created_videos = []
         import time
-        output_filename = f"quote_video_{int(time.time())}.mp4"
-        output_path = os.path.join(output_dir, output_filename)
+        timestamp = int(time.time())
         
-        # Create the video
-        creation_result = create_quote_video_tool(
-            video_result["video_url"],
-            quote,
-            output_path
-        )
+        for i, video_result in enumerate(video_results):
+            output_filename = f"quote_video_v{i+1}_{timestamp}.mp4"
+            output_path = os.path.join(output_dir, output_filename)
+            
+            creation_result = create_quote_video_tool(
+                video_result["video_url"],
+                quote,
+                output_path,
+                audio_path if add_voice else None
+            )
+            
+            if creation_result["success"]:
+                created_videos.append(creation_result["output_path"])
+                status_log.append(f"   ✅ Variation {i+1} created!")
+            else:
+                status_log.append(f"   ⚠️ Variation {i+1} failed")
         
-        if not creation_result["success"]:
-            status_log.append(f"   ❌ Video creation failed: {creation_result['message']}")
-            status_log.append("\n📺 **PREVIEW MODE:**")
-            status_log.append("   Showing background video preview instead")
-            return "\n".join(status_log), video_result["video_url"], None
+        if not created_videos:
+            status_log.append("\n❌ All video creations failed")
+            return "\n".join(status_log), video_results[0]["video_url"] if video_results else None, []
         
-        status_log.append(f"   ✅ Video created successfully!\n")
+        status_log.append("")
         
-        # STEP 5: MCP Server integration status
+        # STEP 6: MCP Server integration status
         status_log.append("🔗 **MCP SERVER STATUS:**")
         if mcp_enabled:
             status_log.append("   ✅ Connected to: abidlabs-mcp-tools.hf.space")
-            status_log.append("   ✅ Additional MCP tools available")
         else:
             status_log.append("   ⚠️ MCP server connection pending")
         status_log.append("")
         
-        # STEP 6: Success!
+        # STEP 7: Success!
         status_log.append("✨ **PIPELINE COMPLETE!**")
-        status_log.append(f"   📱 Original video: {video_result['pexels_url']}")
-        status_log.append(f"   🎬 Final video ready for download!")
+        status_log.append(f"   🎬 Created {len(created_videos)} video variations")
+        if add_voice:
+            status_log.append(f"   🎤 With AI voice narration")
+        status_log.append(f"   📥 Choose your favorite and download!")
         
         final_status = "\n".join(status_log)
-        return final_status, video_result["video_url"], creation_result["output_path"]
+        return final_status, video_results[0]["video_url"] if video_results else None, created_videos
     
     except Exception as e:
         status_log.append(f"\n❌ Pipeline error: {str(e)}")
-        return "\n".join(status_log), None, None
+        return "\n".join(status_log), None, []
 
-def fallback_pipeline(niche, style):
+def fallback_pipeline(niche, style, num_variations=3, add_voice=True):
     """Fallback pipeline if MCP agent fails"""
     status_log = []
     status_log.append("🔄 **FALLBACK MODE (Direct Tool Execution)**\n")
@@ -460,54 +556,85 @@ def fallback_pipeline(niche, style):
     quote = generate_quote_tool(niche, style)
     
     if "Error" in quote:
-        return "\n".join(status_log) + f"\n❌ {quote}", None, None
+        return "\n".join(status_log) + f"\n❌ {quote}", None, []
     
-    status_log.append(f"   ✅ Quote: \"{quote}\"\n")
+    status_log.append(f"   ✅ Quote generated\n")
     
-    # Search video
-    status_log.append("🔍 Searching for video...")
-    video_result = search_pexels_video_tool(style, niche)
+    # Generate voice if requested
+    audio_path = None
+    if add_voice:
+        status_log.append("🎤 Creating voice narration...")
+        audio_dir = "/tmp/quote_audio"
+        os.makedirs(audio_dir, exist_ok=True)
+        
+        import time
+        audio_filename = f"narration_{int(time.time())}.mp3"
+        audio_path = os.path.join(audio_dir, audio_filename)
+        
+        voice_result = generate_voice_narration_tool(quote, audio_path)
+        if voice_result["success"]:
+            status_log.append(f"   ✅ Voice created\n")
+        else:
+            audio_path = None
+            status_log.append(f"   ⚠️ Voice failed\n")
     
-    if not video_result["success"]:
-        return "\n".join(status_log) + f"\n❌ {video_result.get('error', 'Failed')}", None, None
+    # Search videos
+    status_log.append(f"🔍 Searching for {num_variations} videos...")
+    video_results = []
+    for i in range(num_variations):
+        video_result = search_pexels_video_tool(style, niche)
+        if video_result["success"]:
+            video_results.append(video_result)
+    
+    if not video_results:
+        return "\n".join(status_log) + "\n❌ No videos found", None, []
     
-    status_log.append(f"   ✅ Found: {video_result['search_query']}\n")
+    status_log.append(f"   ✅ Found {len(video_results)} videos\n")
     
-    # Create video
-    status_log.append("🎬 Creating final video...")
+    # Create videos
+    status_log.append("🎬 Creating videos...")
     output_dir = "/tmp/quote_videos"
     os.makedirs(output_dir, exist_ok=True)
     
     import time
-    output_filename = f"quote_video_{int(time.time())}.mp4"
-    output_path = os.path.join(output_dir, output_filename)
+    timestamp = int(time.time())
+    created_videos = []
     
-    creation_result = create_quote_video_tool(
-        video_result["video_url"],
-        quote,
-        output_path
-    )
+    for i, video_result in enumerate(video_results):
+        output_filename = f"quote_video_v{i+1}_{timestamp}.mp4"
+        output_path = os.path.join(output_dir, output_filename)
+        
+        creation_result = create_quote_video_tool(
+            video_result["video_url"],
+            quote,
+            output_path,
+            audio_path if add_voice else None
+        )
+        
+        if creation_result["success"]:
+            created_videos.append(creation_result["output_path"])
     
-    if not creation_result["success"]:
-        status_log.append(f"   ❌ {creation_result['message']}")
-        return "\n".join(status_log), video_result["video_url"], None
+    if not created_videos:
+        return "\n".join(status_log) + "\n❌ Video creation failed", video_results[0]["video_url"] if video_results else None, []
     
-    status_log.append("   ✅ Video created!\n")
+    status_log.append(f"   ✅ Created {len(created_videos)} videos!\n")
     status_log.append("🎬 **COMPLETE!**")
     
-    return "\n".join(status_log), video_result["video_url"], creation_result["output_path"]
+    return "\n".join(status_log), video_results[0]["video_url"] if video_results else None, created_videos
 
 # Gradio Interface
 with gr.Blocks(title="AIQuoteClipGenerator - MCP Edition", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # 🎬 AIQuoteClipGenerator
-    ### MCP-Powered Autonomous AI Agent
+    ### MCP-Powered Autonomous AI Agent with Voice Narration
     
     **MCP Integration Features:**
     - 🔗 **MCP Server:** Connected to smolagents framework
-    - 🛠️ **Custom Tools:** Quote generation + Video search + Video creation
+    - 🛠️ **4 Custom MCP Tools:** Quote generation + Video search + Voice narration + Video creation
     - 🤖 **Agent Reasoning:** Autonomous task execution
     - ⚡ **Tool Orchestration:** Intelligent pipeline management
+    - 🎤 **ElevenLabs Voice:** AI narration for videos
+    - 🎨 **Multiple Variations:** Get 3 different video styles
     """)
     
     with gr.Row():
@@ -539,6 +666,21 @@ with gr.Blocks(title="AIQuoteClipGenerator - MCP Edition", theme=gr.themes.Soft(
                 value="Cinematic"
             )
             
+            num_variations = gr.Slider(
+                minimum=1,
+                maximum=5,
+                value=3,
+                step=1,
+                label="🎬 Number of Video Variations",
+                info="Generate multiple versions to choose from"
+            )
+            
+            add_voice = gr.Checkbox(
+                value=True,
+                label="🎤 Add Voice Narration (ElevenLabs)",
+                info="AI voice will read the quote"
+            )
+            
             generate_btn = gr.Button("🤖 Run MCP Agent", variant="primary", size="lg")
         
         with gr.Column():
@@ -550,30 +692,45 @@ with gr.Blocks(title="AIQuoteClipGenerator - MCP Edition", theme=gr.themes.Soft(
             gr.Markdown("### 🎥 Background Video Preview")
             preview_video = gr.Video(label="Original Pexels Video")
         
-        with gr.Column():
-            gr.Markdown("### ✨ Final Quote Video")
-            final_video = gr.Video(label="Download Your Video")
+    with gr.Row():
+        gr.Markdown("### ✨ Your Quote Videos (Pick Your Favorite!)")
+    
+    with gr.Row():
+        video_gallery = gr.Gallery(
+            label="Video Variations",
+            show_label=False,
+            elem_id="gallery",
+            columns=3,
+            rows=2,
+            height="auto",
+            object_fit="contain"
+        )
     
     gr.Markdown("""
     ---
+    ### ✨ NEW FEATURES!
+    - 🎤 **ElevenLabs Voice Narration** - AI voice reads your quotes
+    - 🎨 **Multiple Variations** - Get 3-5 different videos to choose from
+    - ✅ **4 MCP Tools** - Quote, Video Search, Voice, Video Creation
+    
     ### ✨ MCP Implementation
     - ✅ **smolagents Framework** - Proper MCP integration
-    - ✅ **Custom MCP Tools** - Quote generation, video search & video creation
+    - ✅ **Custom MCP Tools** - 4 tools working autonomously
     - ✅ **CodeAgent** - Autonomous reasoning and execution
     - ✅ **MCP Client** - Connected to external MCP servers
-    - ✅ **MoviePy Processing** - Text overlay with professional styling
-    - 🚧 **ElevenLabs Integration** - Voice narration (future)
+    - ✅ **MoviePy + PIL** - Professional text overlay
+    - ✅ **ElevenLabs** - AI voice narration
     
     ### 🏆 Hackathon: MCP 1st Birthday
     **Track:** Track 2 - MCP in Action  
     **Category:** Productivity Tools  
-    **Built with:** Gradio + smolagents + OpenAI + Pexels + MoviePy + MCP
+    **Built with:** Gradio + smolagents + OpenAI + Pexels + ElevenLabs + MoviePy + MCP
     """)
     
     generate_btn.click(
         mcp_agent_pipeline, 
-        inputs=[niche, style], 
-        outputs=[output, preview_video, final_video]
+        inputs=[niche, style, num_variations, add_voice], 
+        outputs=[output, preview_video, video_gallery]
     )
 
 if __name__ == "__main__":

requirements.txt

@@ -9,4 +9,5 @@ imageio-ffmpeg
 decorator
 proglog
 numpy
-Pillow
+Pillow
+elevenlabs