update grid

app.py

@@ -1,24 +1,31 @@
-import gradio as gr
 import os
-import requests
 import random
-import tempfile
 import base64
+import tempfile
 from typing import Tuple, List, Dict, Any
 
+import gradio as gr
+import requests
 from openai import OpenAI
 from smolagents import CodeAgent, MCPClient, tool
 from huggingface_hub import InferenceClient
 from elevenlabs import ElevenLabs, VoiceSettings
 
-# Import our Gemini + OpenAI hybrid quote generator
 from quote_generator_gemini import HybridQuoteGenerator
 
-# ==== CLIENTS / GLOBALS =======================================================
+# =============================================================================
+# GLOBAL CLIENTS / CONFIG
+# =============================================================================
 
 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"))
+
+# ElevenLabs client (optional)
+try:
+    elevenlabs_client = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))
+except Exception as e:
+    print(f"ElevenLabs init warning: {e}")
+    elevenlabs_client = None
 
 # Hybrid quote generator (Gemini primary, OpenAI fallback)
 hybrid_quote_generator = HybridQuoteGenerator(
@@ -37,11 +44,13 @@ except Exception as e:
 # Modal endpoint for fast video rendering
 MODAL_ENDPOINT_URL = os.getenv("MODAL_ENDPOINT_URL")
 
-# ==== CONTEXT ENGINEERING: PERSONAS & TRENDS ==================================
 
+# =============================================================================
+# CONTEXT ENGINEERING: PERSONA + TRENDS
+# =============================================================================
 
 def get_persona_instruction(persona: str) -> str:
-    """Return a short style instruction for the selected persona."""
+    """Short style instruction for the selected persona."""
     persona = (persona or "").lower()
     if persona == "coach":
         return (
@@ -68,8 +77,7 @@ def get_persona_instruction(persona: str) -> str:
 
 def get_trend_insights(niche: str) -> Dict[str, Any]:
     """
-    Lightweight 'RAG' over static trend knowledge per niche.
-    Returns topics + hooks + a short summary string.
+    Lightweight 'trend RAG' – returns niche-specific themes & hooks.
     """
     niche = niche or "Motivation"
 
@@ -77,9 +85,8 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
         "Motivation": {
             "label": "soft life vs discipline era",
             "summary": (
-                "Right now motivational content leans into 'soft life' aesthetics while still "
-                "talking about discipline, systems, and quiet consistency. People want ambition "
-                "without burnout and routines that feel gentle but effective."
+                "Motivational content leans into 'soft life' aesthetics while still "
+                "talking about discipline, systems, and quiet consistency."
             ),
             "topics": [
                 {
@@ -92,53 +99,49 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
                 },
                 {
                     "topic": "Reset Routine Hacks",
-                    "hook": "Feeling stuck? Here’s a 10-minute reset that gets you moving again.",
-                },
-                {
-                    "topic": "Motivation in 60 Seconds",
-                    "hook": "Your 60-second reminder that small moves count more than perfect plans.",
+                    "hook": "A 10-minute reset to get you unstuck.",
                 },
             ],
         },
         "Business/Entrepreneurship": {
             "label": "one-person brands & slow growth",
             "summary": (
-                "Founders are tired of hustle theatre. Trending content focuses on one-person "
-                "brands, slow compounding, transparent revenue, and honest behind-the-scenes."
+                "Founders are tired of hustle theatre. Trending content focuses on "
+                "one-person brands, slow compounding, and honest behind-the-scenes."
             ),
             "topics": [
                 {
                     "topic": "Build in Public Moments",
-                    "hook": "Here’s the part of building nobody shows you—but everyone feels.",
+                    "hook": "Here’s the part of building nobody shows—but everyone feels.",
                 },
                 {
                     "topic": "Tiny Experiments",
-                    "hook": "Instead of a 5-year plan, here’s one experiment you can run this week.",
+                    "hook": "One small experiment you can run this week instead of a 5-year plan.",
                 },
             ],
         },
         "Fitness": {
             "label": "sustainable glow-up",
             "summary": (
-                "Fitness trends lean toward sustainable glow-ups: lower-impact routines, walking, "
-                "strength, and realistic body expectations."
+                "Fitness trends lean toward sustainable glow-ups: walking, strength, "
+                "and realistic body expectations."
             ),
             "topics": [
                 {
                     "topic": "Gentle Discipline Workouts",
-                    "hook": "A workout routine for the days you ‘don’t feel like it’ but still care.",
+                    "hook": "A routine for the days you ‘don’t feel like it’ but still care.",
                 },
                 {
                     "topic": "Slow Glow-Up",
-                    "hook": "Your glow-up doesn’t need to be loud. Here’s the quiet version.",
+                    "hook": "The quiet glow-up that happens when you stop quitting.",
                 },
             ],
         },
         "Mindfulness": {
             "label": "nervous system & soft resets",
             "summary": (
-                "Mindfulness content is shifting toward nervous system regulation, tiny resets, "
-                "and practical grounding instead of abstract spirituality."
+                "Mindfulness content is shifting toward nervous system regulation, tiny "
+                "resets, and practical grounding."
             ),
             "topics": [
                 {
@@ -154,13 +157,13 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
         "Stoicism": {
             "label": "quiet strength",
             "summary": (
-                "Stoic content focuses on quiet strength, emotional regulation, and not reacting "
-                "to every notification, comment, or impulse."
+                "Stoic content focuses on quiet strength, emotional regulation, and not "
+                "reacting to every notification, comment, or impulse."
             ),
             "topics": [
                 {
                     "topic": "Reaction Discipline",
-                    "hook": "You can’t control people—but you can control the delay before you answer.",
+                    "hook": "You can’t control people—but you can control the pause before you answer.",
                 },
                 {
                     "topic": "Modern Stoic Moments",
@@ -171,13 +174,13 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
         "Leadership": {
             "label": "servant leadership & clarity",
             "summary": (
-                "Leadership trends highlight servant leadership, psychological safety, and clear, "
-                "simple direction instead of corporate buzzwords."
+                "Leadership trends highlight servant leadership, psychological safety, "
+                "and simple, clear direction."
             ),
             "topics": [
                 {
                     "topic": "Clarity Over Charisma",
-                    "hook": "People don’t need a hero. They need one simple, clear next step.",
+                    "hook": "People don’t need a hero. They need one clear next step.",
                 },
                 {
                     "topic": "Leader as Mirror",
@@ -188,8 +191,8 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
         "Love & Relationships": {
             "label": "self-worth & secure attachment",
             "summary": (
-                "Relationship content leans into self-worth, boundaries, and secure attachment—not "
-                "only romance but choosing emotionally safe people."
+                "Relationship content leans into self-worth, boundaries, and secure "
+                "attachment—not just romance but emotional safety."
             ),
             "topics": [
                 {
@@ -207,8 +210,8 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
     default = {
         "label": "modern glow-up & gentle discipline",
         "summary": (
-            "Short-form content leans into gentle discipline, realistic routines, and soft glow-ups "
-            "instead of extreme hustle or perfection."
+            "Short-form content leans into gentle discipline, realistic routines, "
+            "and soft glow-ups instead of extreme hustle."
         ),
         "topics": [
             {
@@ -221,8 +224,9 @@ def get_trend_insights(niche: str) -> Dict[str, Any]:
     return trends.get(niche, default)
 
 
-# ==== TOOLS (FOR AGENT + DIRECT USE) ==========================================
-
+# =============================================================================
+# TOOLS
+# =============================================================================
 
 @tool
 def generate_quote_tool(niche: str, style: str, persona: str) -> str:
@@ -238,7 +242,6 @@ def generate_quote_tool(niche: str, style: str, persona: str) -> str:
     Returns:
         A unique quote string.
     """
-    # Combine style + persona into a richer conditioning string
     persona_instruction = get_persona_instruction(persona)
     combined_style = f"{style} | persona={persona} | tone={persona_instruction}"
 
@@ -276,14 +279,8 @@ def search_pexels_video_tool(style: str, niche: str, trend_label: str = "") -> d
         trend_label: Short description of the current trend for this niche.
 
     Returns:
-        Dictionary with:
-            - success: bool
-            - video_url: URL of selected MP4
-            - search_query: query used
-            - pexels_url: original Pexels page
-            - error: message if any
+        Dictionary with success, video_url, search_query, pexels_url, error (if any).
     """
-    # Base search strategy
     base_queries = {
         "Motivation": {
             "Cinematic": ["running sunrise", "cliff sunrise", "city at dawn"],
@@ -339,7 +336,6 @@ def search_pexels_video_tool(style: str, niche: str, trend_label: str = "") -> d
     niche_map = base_queries.get(niche, base_queries["Motivation"])
     queries = niche_map.get(style, niche_map["Cinematic"])
 
-    # Light tweak: if trend label contains certain words, bias the query
     trend_label_lower = (trend_label or "").lower()
     if "soft life" in trend_label_lower:
         queries = queries + ["soft life aesthetic", "cozy morning light"]
@@ -359,9 +355,7 @@ def search_pexels_video_tool(style: str, niche: str, trend_label: str = "") -> d
             video_files = video.get("video_files", [])
 
             portrait_videos = [
-                vf
-                for vf in video_files
-                if vf.get("width", 0) < vf.get("height", 0)
+                vf for vf in video_files if vf.get("width", 0) < vf.get("height", 0)
             ]
 
             if portrait_videos:
@@ -413,8 +407,8 @@ def create_quote_video_tool(
         video_url: Direct URL to a Pexels MP4 file.
         quote_text: The quote text to overlay on the video.
         output_path: Local path where the MP4 should be saved.
-        audio_b64: Optional base64-encoded audio for narration (mp3).
-        text_style: Layout style for quote text (classic_center, lower_third_serif, typewriter_top).
+        audio_b64: Base64-encoded audio for narration.
+        text_style: Layout style for quote text.
 
     Returns:
         Dictionary with success flag, message, and output_path if successful.
@@ -475,8 +469,9 @@ def create_quote_video_tool(
         }
 
 
-# ==== OPTIONAL: AGENT INITIALIZATION (MCP-FLAVORED) ===========================
-
+# =============================================================================
+# AGENT (MCP-FLAVORED)
+# =============================================================================
 
 def initialize_agent():
     """Initialize the CodeAgent with MCP capabilities (if available)."""
@@ -516,8 +511,10 @@ def initialize_agent():
 
 agent, agent_error = initialize_agent()
 
-# ==== VOICE GENERATION (OpenAI explanation + ElevenLabs TTS) ==================
 
+# =============================================================================
+# VOICE GENERATION (OpenAI commentary + ElevenLabs TTS)
+# =============================================================================
 
 def get_voice_config(voice_profile: str) -> Tuple[str, VoiceSettings]:
     """
@@ -528,7 +525,7 @@ def get_voice_config(voice_profile: str) -> Tuple[str, VoiceSettings]:
     # Calm female (Rachel)
     if "rachel" in vp or "female" in vp:
         return (
-            "21m00Tcm4TlvDq8ikWAM",  # Rachel (from ElevenLabs docs)
+            "21m00Tcm4TlvDq8ikWAM",  # Rachel
             VoiceSettings(
                 stability=0.5,
                 similarity_boost=0.9,
@@ -537,7 +534,7 @@ def get_voice_config(voice_profile: str) -> Tuple[str, VoiceSettings]:
             ),
         )
 
-    # Warm male (Adam)
+    # Warm male (Adam) – default
     return (
         "pNInz6obpgDQGcFmaJgB",  # Adam
         VoiceSettings(
@@ -558,7 +555,6 @@ def generate_voice_commentary(
 ) -> Tuple[str, str]:
     """
     Generate a short explanatory commentary + ElevenLabs audio (as base64).
-    Voice is always generated if ElevenLabs is available.
 
     Returns:
         (commentary_text, audio_b64) – audio_b64 may be "" if error.
@@ -566,7 +562,6 @@ def generate_voice_commentary(
     if not elevenlabs_client:
         return "", ""
 
-    # 1) Use OpenAI to generate brief commentary
     persona_instruction = get_persona_instruction(persona)
     prompt = f"""
 You are creating a short voice-over commentary for a TikTok/Instagram quote video.
@@ -590,6 +585,7 @@ Requirements:
 Return ONLY the commentary text, nothing else.
 """
 
+    # 1) Commentary via OpenAI
     try:
         completion = openai_client.chat.completions.create(
             model="gpt-4o-mini",
@@ -600,7 +596,6 @@ Return ONLY the commentary text, nothing else.
             max_tokens=120,
             temperature=0.7,
         )
-
         commentary = completion.choices[0].message.content.strip()
     except Exception as e:
         print(f"⚠️ Error generating commentary text: {e}")
@@ -609,24 +604,23 @@ Return ONLY the commentary text, nothing else.
     # 2) ElevenLabs TTS
     try:
         voice_id, voice_settings = get_voice_config(voice_profile)
-
         audio_stream = elevenlabs_client.text_to_speech.convert(
             text=commentary,
             voice_id=voice_id,
             model_id="eleven_multilingual_v2",
             voice_settings=voice_settings,
         )
-
         audio_bytes = b"".join(chunk for chunk in audio_stream)
         audio_b64 = base64.b64encode(audio_bytes).decode("utf-8")
         return commentary, audio_b64
     except Exception as e:
         print(f"⚠️ Error generating ElevenLabs audio: {e}")
-        return commentary, ""  # keep commentary for logs even if audio fails
-
+        return commentary, ""
 
-# ==== PIPELINE (ADVANCED AGENT-LIKE FLOW) =====================================
 
+# =============================================================================
+# PIPELINE (MCP-STYLE)
+# =============================================================================
 
 def mcp_agent_pipeline(
     niche: str,
@@ -638,10 +632,10 @@ def mcp_agent_pipeline(
 ) -> Tuple[str, List[str]]:
     """
     MCP-flavored autonomous pipeline with:
-    - Context engineering (persona, trends)
-    - Trend-informed 'RAG' context injection
+    - Context engineering (niche + persona + trends)
+    - Trend-informed context
     - Quote generation via hybrid Gemini/OpenAI
-    - ElevenLabs narration (always on if available)
+    - ElevenLabs narration
     - Modal-based video creation (1–3 variations)
     """
 
@@ -652,7 +646,7 @@ def mcp_agent_pipeline(
         status_log.append(f"⚠️ Agent initialization failed: {agent_error}")
         status_log.append("   Falling back to direct tool execution.\n")
 
-    # STEP 0: Context & trends
+    # Step 0: context & trends
     status_log.append("🧩 **Step 0 – Building context**")
     status_log.append(f"   • Niche: `{niche}`")
     status_log.append(f"   • Visual style: `{style}`")
@@ -665,20 +659,19 @@ def mcp_agent_pipeline(
     trend_summary = trend_info.get("summary", "")
     topics_for_log = ", ".join(t["topic"] for t in trend_info.get("topics", [])[:3])
 
-    status_log.append("📈 **Step 1 – Trend-aware context (mini-RAG)**")
+    status_log.append("📈 **Step 1 – Trend-aware context**")
     status_log.append(f"   • Trend theme: {trend_label}")
     status_log.append(f"   • Topics: {topics_for_log}")
     status_log.append(f"   • Summary: {trend_summary}\n")
 
-    # Simple “fusion score” heuristic
     fusion_score = random.randint(78, 97)
     status_log.append(
         f"🎯 **Context Fusion Score:** {fusion_score}/100 "
         "(niche + trend + persona alignment)\n"
     )
 
-    # STEP 2: Generate quote
-    status_log.append("🧠 **Step 2 – Generating quote (Gemini + variety tracking)**")
+    # Step 2: quote
+    status_log.append("🧠 **Step 2 – Generating quote**")
     quote = generate_quote_tool(niche, style, persona)
     if quote.startswith("Error"):
         status_log.append(f"   ❌ Quote generation error: {quote}")
@@ -687,8 +680,8 @@ def mcp_agent_pipeline(
     preview = quote if len(quote) <= 140 else quote[:140] + "..."
     status_log.append(f"   ✅ Quote: “{preview}”\n")
 
-    # STEP 3: Voice commentary (always attempted)
-    status_log.append("🔊 **Step 3 – Generating voice-over explanation (OpenAI + ElevenLabs)**")
+    # Step 3: voice commentary
+    status_log.append("🔊 **Step 3 – Generating voice-over (OpenAI + ElevenLabs)**")
     commentary, audio_b64 = generate_voice_commentary(
         quote_text=quote,
         niche=niche,
@@ -697,13 +690,13 @@ def mcp_agent_pipeline(
         voice_profile=voice_profile,
     )
     if audio_b64:
-        status_log.append("   ✅ Voice-over created and encoded as base64")
+        status_log.append("   ✅ Voice-over created")
     else:
         status_log.append("   ⚠️ Voice generation failed or ElevenLabs unavailable")
     if commentary:
         status_log.append(f"   📝 Commentary preview: {commentary[:120]}...\n")
 
-    # STEP 4: Search Pexels videos
+    # Step 4: Pexels videos
     status_log.append("🎥 **Step 4 – Searching Pexels for background videos**")
     status_log.append(f"   Target variations: {num_variations}\n")
 
@@ -713,11 +706,12 @@ def mcp_agent_pipeline(
         if vr.get("success"):
             video_results.append(vr)
             status_log.append(
-                f"   ✅ Variation {i+1}: query=`{vr['search_query']}`  url={vr['pexels_url']}"
+                f"   ✅ Variation {i+1}: query=`{vr['search_query']}` url={vr['pexels_url']}"
             )
         else:
             status_log.append(
-                f"   ⚠️ Variation {i+1} video search failed: {vr.get('error', 'unknown error')}"
+                f"   ⚠️ Variation {i+1} video search failed: "
+                f"{vr.get('error', 'unknown error')}"
             )
 
     if not video_results:
@@ -726,7 +720,7 @@ def mcp_agent_pipeline(
 
     status_log.append("")
 
-    # STEP 5: Create videos via Modal
+    # Step 5: Modal rendering
     status_log.append("🎬 **Step 5 – Rendering quote videos on Modal**")
     output_dir = "/tmp/quote_videos"
     gallery_dir = "/data/gallery_videos"
@@ -737,7 +731,6 @@ def mcp_agent_pipeline(
     import shutil
 
     timestamp = int(time.time())
-
     created_videos: List[str] = []
 
     for i, vr in enumerate(video_results):
@@ -754,9 +747,8 @@ def mcp_agent_pipeline(
 
         if creation_result.get("success"):
             created_videos.append(out_path)
-            status_log.append(f"   ✅ Variation {i+1} rendered successfully")
+            status_log.append(f"   ✅ Variation {i+1} rendered")
 
-            # Copy to gallery (we keep ALL; scrolling handled by Gradio gallery)
             gallery_filename = f"gallery_{timestamp}_v{i+1}.mp4"
             gallery_path = os.path.join(gallery_dir, gallery_filename)
             try:
@@ -773,12 +765,12 @@ def mcp_agent_pipeline(
         status_log.append("\n❌ All video renderings failed.")
         return "\n".join(status_log), []
 
-    # STEP 6: Wrap up
+    # Wrap-up
     status_log.append("\n🔗 **Integrations used:**")
-    status_log.append("   • Gemini (via HybridQuoteGenerator) – quote + variety tracking")
-    status_log.append("   • OpenAI – spoken-style explanation for voice-over")
+    status_log.append("   • Gemini – quote + variety tracking")
+    status_log.append("   • OpenAI – spoken-style commentary")
     status_log.append("   • ElevenLabs – voice narration")
-    status_log.append("   • Pexels – background stock video search")
+    status_log.append("   • Pexels – stock video search")
     status_log.append("   • Modal – fast video rendering")
     if mcp_enabled:
         status_log.append("   • MCP server – available for extended tools")
@@ -789,19 +781,19 @@ def mcp_agent_pipeline(
     return "\n".join(status_log), created_videos
 
 
-# ==== GALLERY UTIL (SCROLLABLE, KEEPS ALL) ====================================
-
+# =============================================================================
+# GALLERY (SCROLLABLE GRID)
+# =============================================================================
 
 def load_gallery_videos() -> List[str]:
     """
     Load all videos from persistent gallery folder (sorted newest → oldest).
-    Gradio's Gallery will handle scrolling.
+    Gradio's Gallery will handle scrolling in a grid layout.
     """
     gallery_output_dir = "/data/gallery_videos"
     os.makedirs(gallery_output_dir, exist_ok=True)
 
     import glob
-
     existing_videos = sorted(
         glob.glob(f"{gallery_output_dir}/*.mp4"),
         key=os.path.getmtime,
@@ -811,7 +803,9 @@ def load_gallery_videos() -> List[str]:
     return existing_videos
 
 
-# ==== GRADIO UI ===============================================================
+# =============================================================================
+# GRADIO UI
+# =============================================================================
 
 with gr.Blocks(
     title="AIQuoteClipGenerator - MCP + Gemini Edition",
@@ -820,23 +814,24 @@ with gr.Blocks(
     gr.Markdown(
         """
     # 🎬 AIQuoteClipGenerator
-    ### MCP-flavored agent • Gemini + OpenAI + ElevenLabs + Modal
+    ### MCP-style agent • Gemini + OpenAI + ElevenLabs + Modal
 
     An autonomous mini-studio that generates trend-aware quote videos with voice-over,
     cinematic stock footage, and MCP-style agent reasoning.
     """
     )
 
+    # Scrollable IG-style gallery
     with gr.Accordion("📸 Example Gallery – All Generated Videos", open=True):
         gr.Markdown("Scroll to explore all the clips you've generated so far.")
         gallery = gr.Gallery(
             label=None,
-            elem_id="gallery",
             show_label=False,
-            columns=[3],
-            rows=[2],
-            height=540,
-            object_fit="cover",
+            columns=3,          # 3 per row → IG grid
+            rows=2,             # ~2 rows visible before scroll
+            height=540,         # scrolling kicks in after this
+            object_fit="cover", # crop previews to fill tiles
+            allow_preview=True,
             preview=True,
         )
 
@@ -862,13 +857,7 @@ with gr.Blocks(
             )
 
             style = gr.Dropdown(
-                choices=[
-                    "Cinematic",
-                    "Nature",
-                    "Urban",
-                    "Minimal",
-                    "Abstract",
-                ],
+                choices=["Cinematic", "Nature", "Urban", "Minimal", "Abstract"],
                 label="🎨 Visual Style",
                 value="Cinematic",
             )
@@ -880,11 +869,7 @@ with gr.Blocks(
             )
 
             text_style = gr.Dropdown(
-                choices=[
-                    "classic_center",
-                    "lower_third_serif",
-                    "typewriter_top",
-                ],
+                choices=["classic_center", "lower_third_serif", "typewriter_top"],
                 label="🖋 Text Layout Style",
                 value="classic_center",
             )
@@ -908,7 +893,8 @@ with gr.Blocks(
             )
 
             generate_btn = gr.Button(
-                "🤖 Run Agent Pipeline", variant="primary"
+                "🤖 Run Agent Pipeline",
+                variant="primary",
             )
 
         with gr.Column():
@@ -959,13 +945,7 @@ with gr.Blocks(
 
         gallery_vids = load_gallery_videos()
 
-        return [
-            status,
-            v1,
-            v2,
-            v3,
-            gallery_vids,
-        ]
+        return status, v1, v2, v3, gallery_vids
 
     generate_btn.click(
         process_and_display,