Update app.py

app.py

@@ -1,4 +1,4 @@
-import os, tempfile, io, math, time, threading, re, random
+import os, tempfile, io, math, time, threading, re, random, json
 import numpy as np
 import cv2
 import gradio as gr
@@ -14,52 +14,83 @@ if hf_token:
         print("✅ HF ready")
     except Exception as e: print(f"⚠️ HF: {e}")
 
+# ── TEMPLATE STORAGE ──────────────────────────────────────────────
+TEMPLATES_FILE = "saved_templates.json"
+
+def load_templates():
+    if os.path.exists(TEMPLATES_FILE):
+        try:
+            with open(TEMPLATES_FILE, "r") as f:
+                return json.load(f)
+        except: pass
+    return {}
+
+def save_template(name, style, language, duration, caption, add_aud, add_cap):
+    templates = load_templates()
+    templates[name] = {
+        "style": style, "language": language, "duration": duration,
+        "caption": caption, "add_audio": add_aud, "add_captions": add_cap,
+        "created": time.strftime("%Y-%m-%d %H:%M")
+    }
+    with open(TEMPLATES_FILE, "w") as f:
+        json.dump(templates, f, indent=2)
+    return f"✅ Template '{name}' saved!", list(templates.keys())
+
+def get_template_names():
+    return list(load_templates().keys())
+
+def load_template(name):
+    templates = load_templates()
+    if name in templates:
+        t = templates[name]
+        return t["style"], t["language"], t["duration"], t["caption"], t["add_audio"], t["add_captions"]
+    return "Premium", "English", 6, "", True, True
+
+def delete_template(name):
+    templates = load_templates()
+    if name in templates:
+        del templates[name]
+        with open(TEMPLATES_FILE, "w") as f:
+            json.dump(templates, f, indent=2)
+        return f"🗑️ Template '{name}' deleted!", list(templates.keys())
+    return "Template not found!", list(templates.keys())
+
+def export_template(name):
+    templates = load_templates()
+    if name in templates:
+        out = f"template_{name.replace(' ','_')}.json"
+        with open(out, "w") as f:
+            json.dump({name: templates[name]}, f, indent=2)
+        return out
+    return None
+
 # ══════════════════════════════════════════════════════════════════
-#  AUTO-DETECT from image (color + HF classifier)
+#  AUTO-DETECT
 # ══════════════════════════════════════════════════════════════════
 def auto_detect(pil_image, user_caption=""):
-    """
-    1. Try HF image classification
-    2. Fallback: dominant color + aspect ratio heuristics
-    Returns (category, auto_prompt, auto_caption_hint)
-    """
     category = "Product/Other"
     label    = ""
-
-    # Try HF zero-shot image classification
     if hf_client:
         try:
-            buf = io.BytesIO(); pil_image.save(buf,format="JPEG",quality=85)
-            result = hf_client.image_classification(
-                image=buf.getvalue(),
-                model="google/vit-base-patch16-224",
-            )
+            buf = io.BytesIO(); pil_image.save(buf, format="JPEG", quality=85)
+            result = hf_client.image_classification(image=buf.getvalue(), model="google/vit-base-patch16-224")
             if result:
                 label = result[0].get("label","").lower()
-                print(f"  🔍 HF label: {label}")
         except Exception as e:
             print(f"  ⚠️ classifier skip: {e}")
 
-    # Map HF label → our category
     label_map = {
-        "shoe":    "Fashion", "sneaker": "Fashion", "boot":    "Fashion",
-        "dress":   "Fashion", "shirt":   "Fashion", "jacket":  "Fashion",
-        "jean":    "Fashion", "sandal":  "Fashion", "bag":     "Fashion",
-        "pizza":   "Food",    "burger":  "Food",    "cake":    "Food",
-        "food":    "Food",    "coffee":  "Food",    "sushi":   "Food",
-        "laptop":  "Tech",    "phone":   "Tech",    "camera":  "Tech",
-        "keyboard":"Tech",    "monitor": "Tech",    "tablet":  "Tech",
-        "lipstick":"Beauty",  "cream":   "Beauty",  "perfume": "Beauty",
-        "cosmetic":"Beauty",  "makeup":  "Beauty",
-        "dumbbell":"Fitness", "yoga":    "Fitness", "bottle":  "Fitness",
-        "bicycle": "Fitness", "jersey":  "Fitness",
-        "plant":   "Lifestyle","candle": "Lifestyle","chair":  "Lifestyle",
-        "sofa":    "Lifestyle","lamp":   "Lifestyle",
+        "shoe":"Fashion","sneaker":"Fashion","boot":"Fashion","dress":"Fashion",
+        "shirt":"Fashion","jacket":"Fashion","jean":"Fashion","sandal":"Fashion","bag":"Fashion",
+        "pizza":"Food","burger":"Food","cake":"Food","food":"Food","coffee":"Food","sushi":"Food",
+        "laptop":"Tech","phone":"Tech","camera":"Tech","keyboard":"Tech","monitor":"Tech","tablet":"Tech",
+        "lipstick":"Beauty","cream":"Beauty","perfume":"Beauty","cosmetic":"Beauty","makeup":"Beauty",
+        "dumbbell":"Fitness","yoga":"Fitness","bottle":"Fitness","bicycle":"Fitness","jersey":"Fitness",
+        "plant":"Lifestyle","candle":"Lifestyle","chair":"Lifestyle","sofa":"Lifestyle","lamp":"Lifestyle",
     }
     for k,v in label_map.items():
         if k in label: category=v; break
 
-    # Also check user caption
     if category == "Product/Other" and user_caption:
         cap_low = user_caption.lower()
         if any(w in cap_low for w in ["shoe","sneaker","dress","outfit","wear","fashion","style","cloth","kurta"]): category="Fashion"
@@ -69,7 +100,6 @@ def auto_detect(pil_image, user_caption=""):
         elif any(w in cap_low for w in ["gym","fit","workout","protein","yoga","health","sport"]): category="Fitness"
         elif any(w in cap_low for w in ["home","decor","interior","lifestyle","aesthetic","candle"]): category="Lifestyle"
 
-    # Build cinematic prompt from detected category
     prompts = {
         "Fashion":   "cinematic fashion product shot, model wearing outfit, soft studio lighting, slow zoom, luxury feel",
         "Food":      "cinematic food photography, steam rising, dramatic close-up, warm golden lighting, slow reveal",
@@ -81,10 +111,8 @@ def auto_detect(pil_image, user_caption=""):
     }
     auto_prompt = prompts.get(category, prompts["Product/Other"])
     if label: auto_prompt = f"{label} product, {auto_prompt}"
-
     return category, auto_prompt, label
 
-
 # ══════════════════════════════════════════════════════════════════
 #  SMART INSIGHTS
 # ══════════════════════════════════════════════════════════════════
@@ -97,7 +125,6 @@ POSTING_TIMES = {
     "Lifestyle":    {"best":"7:00 PM",  "days":"Thu, Fri, Sat", "slots":["9AM","2PM","7PM"]},
     "Product/Other":{"best":"8:00 PM",  "days":"Tue, Thu, Sat", "slots":["10AM","3PM","8PM"]},
 }
-
 AUDIENCES = {
     "Fashion":      "👗 18-35 yo females · Fashion lovers · Insta scrollers · Trend followers",
     "Food":         "🍕 18-45 · Foodies · Home cooks · Restaurant goers · Food bloggers",
@@ -107,7 +134,6 @@ AUDIENCES = {
     "Lifestyle":    "🌿 22-40 · Aspirational buyers · Aesthetic lovers · Home decor fans",
     "Product/Other":"🛍️ 18-45 · Online shoppers · Deal hunters · Value-conscious buyers",
 }
-
 CAPTIONS = {
     "English": {
         "Premium":   ["✨ {cap} Quality that speaks for itself. 🛒 Shop Now → Link in bio",
@@ -134,7 +160,6 @@ CAPTIONS = {
                       "POV: Naya fav mil gaya 🎉 {cap} Bio mein link!"],
     },
 }
-
 HASHTAGS = {
     "Fashion":   "#Fashion #OOTD #StyleInspo #NewCollection #Trending #ShopNow #Reels",
     "Food":      "#FoodLovers #Foodie #FoodPhotography #Yummy #FoodReels #MustTry",
@@ -151,7 +176,6 @@ def get_insights(category, style, language, cap):
     tmpl = CAPTIONS.get(language, CAPTIONS["English"]).get(style, CAPTIONS["English"]["Premium"])
     ai_cap = random.choice(tmpl).replace("{cap}", clean_cap)
     tags   = HASHTAGS.get(category, HASHTAGS["Product/Other"])
-
     insight = (
         f"📊 SMART INSIGHTS\n"
         f"{'━'*38}\n"
@@ -166,6 +190,79 @@ def get_insights(category, style, language, cap):
     )
     return insight, ai_cap
 
+# ══════════════════════════════════════════════════════════════════
+#  AI EXPLAINER BOT
+# ══════════════════════════════════════════════════════════════════
+BOT_KB = {
+    # Tech questions
+    "how does this work": "🤖 **How it works:**\n1. You upload 1-5 product images\n2. AI auto-detects category (Fashion/Food/Tech etc.) using HuggingFace VIT model\n3. Ken Burns cinematic effect is applied (zoom + pan + color grading)\n4. AI generates captions, hashtags & best posting times\n5. Optional: BGM music + TTS voice overlay added via gTTS\n6. Final video exported as MP4 📱",
+
+    "ken burns": "🎬 **Ken Burns Effect** is a cinematic technique used here:\n- Smooth zoom in/out (scale 1.0→1.06)\n- Gentle pan left/right/up/down\n- Easing curves (cubic) for natural motion\n- Color grading per style (warm tones for Premium, saturated for Energetic)\n- Vignette overlay for cinematic look\n- Fade in/out at start and end\nThis runs 100% locally — no GPU needed! ✅",
+
+    "hf models": "🤗 **HuggingFace AI Chain** (tried in order):\n1. **LTX-2 ⚡** — Lightricks, fastest video gen\n2. **Wan 2.2** — High quality image-to-video\n3. **SVD-XT** — Stable Video Diffusion by Stability AI\n4. **Kling** — KlingTeam's LivePortrait model\n5. **LTX-Video** — Fallback video model\n6. **Ken Burns ✅** — Always works locally!\nEach model gets 50 seconds timeout before trying next.",
+
+    "auto detect": "🔍 **Auto-Detection System:**\n- Uses Google ViT-base-patch16-224 model via HF API\n- Classifies image into 1000 ImageNet categories\n- Maps labels → Fashion/Food/Tech/Beauty/Fitness/Lifestyle\n- Falls back to caption keyword matching if HF unavailable\n- Selects cinematic prompt style based on detected category",
+
+    "captions": "💬 **Caption System:**\n- 3 languages: English / Hindi / Hinglish\n- 3 styles: Premium / Energetic / Fun\n- Uses ffmpeg drawtext filter for overlay\n- Animated fade-in/out effect\n- CTA button ('Shop Now') added automatically\n- Hashtag line auto-appended based on category",
+
+    "audio": "🎵 **Audio Pipeline:**\n- **BGM:** Generated programmatically using numpy (sine waves + kick drum + hi-hat)\n- BPM varies by style: Premium=88, Energetic=126, Fun=104\n- **Voice:** gTTS (Google TTS) narrates your caption\n- Both mixed using ffmpeg amix filter\n- BGM volume=20%, Voice=95%",
+
+    "error": "🔧 **Common Errors & Fixes:**\n- **Gallery Error:** Fixed! Now handles PIL images + numpy arrays safely\n- **HF Timeout:** Models auto-fallback to Ken Burns (always works)\n- **ffmpeg missing:** Install with `apt install ffmpeg`\n- **No video output:** Check if images are valid PNG/JPG\n- **Template not loading:** Templates saved in `saved_templates.json`",
+
+    "template": "💾 **Template System:**\n- Save your settings (style, language, duration, caption) as named templates\n- Load them anytime with one click\n- Export as JSON to share with others\n- Delete templates you no longer need\n- Great for brand consistency across multiple reels!",
+
+    "unique features": "⭐ **Unique Features of this Project:**\n1. 🤖 Multi-model AI chain with auto-fallback\n2. 🔍 Auto category detection from image\n3. 💬 Multilingual captions (Hindi/Hinglish/English)\n4. 🎵 Programmatic BGM generation (no external assets)\n5. 💾 Template save/load/export system\n6. 📊 Smart posting time analytics\n7. 🎬 Custom Ken Burns with style-specific color grading\n8. 🤝 This explainer bot!\n9. 📸 Multi-image → multi-clip merge\n10. 🏷️ Auto-hashtag generation",
+
+    "styles": "🎨 **Style Modes:**\n- **Premium:** Warm tones, 88 BPM, slow elegant zoom, serif feel\n- **Energetic:** High saturation, 126 BPM, dynamic cuts, bold colors\n- **Fun:** Pastel tones, 104 BPM, bouncy motion, playful captions\nEach style affects: color grading, music BPM, caption tone, and CTA color",
+
+    "multi image": "📸 **Multi-Image Reel:**\n- Upload up to 5 images\n- Each image gets its own video clip\n- Duration is split equally across clips\n- All clips merged using ffmpeg concat\n- Result: a smooth multi-product showcase reel!",
+
+    "languages": "🌐 **Language Support:**\n- **English:** Standard captions with emojis\n- **Hindi:** Full Devanagari script captions\n- **Hinglish:** Mixed Hindi-English (very popular on Indian reels)\nFont must support Unicode for Hindi — DejaVu or Liberation used automatically",
+}
+
+def bot_reply(user_msg, history):
+    if not user_msg.strip():
+        return history, ""
+    
+    msg_low = user_msg.lower()
+    reply = None
+    
+    # Match keywords
+    for key, answer in BOT_KB.items():
+        if any(word in msg_low for word in key.split()):
+            reply = answer
+            break
+    
+    # Fuzzy fallbacks
+    if not reply:
+        if any(w in msg_low for w in ["model","ai","hf","huggingface"]): reply = BOT_KB["hf models"]
+        elif any(w in msg_low for w in ["video","cinematic","animation","zoom"]): reply = BOT_KB["ken burns"]
+        elif any(w in msg_low for w in ["detect","category","classify"]): reply = BOT_KB["auto detect"]
+        elif any(w in msg_low for w in ["music","sound","bgm","voice","audio"]): reply = BOT_KB["audio"]
+        elif any(w in msg_low for w in ["caption","text","overlay","subtitle"]): reply = BOT_KB["captions"]
+        elif any(w in msg_low for w in ["save","load","template","export"]): reply = BOT_KB["template"]
+        elif any(w in msg_low for w in ["fix","bug","problem","issue","not working","fail"]): reply = BOT_KB["error"]
+        elif any(w in msg_low for w in ["special","unique","different","best","feature"]): reply = BOT_KB["unique features"]
+        elif any(w in msg_low for w in ["style","premium","energetic","fun"]): reply = BOT_KB["styles"]
+        elif any(w in msg_low for w in ["hindi","hinglish","language","english"]): reply = BOT_KB["languages"]
+        elif any(w in msg_low for w in ["multiple","multi","images","clips","merge"]): reply = BOT_KB["multi image"]
+        elif any(w in msg_low for w in ["hello","hi","hey","namaste","hii"]): reply = "👋 Namaste! Main hun **ReelBot** — tumhara AI guide!\n\nMujhse poocho:\n- 'how does this work'\n- 'ken burns kya hai'\n- 'HF models kaunse hain'\n- 'unique features kya hain'\n- 'error fix kaise karein'\n- 'template kaise use karein'\n\nKoi bhi sawaal pucho! 🚀"
+        else:
+            reply = ("🤔 Hmm, is topic par mujhe exact info nahi mili.\n\n"
+                     "Try asking about:\n"
+                     "• `how does this work` — full pipeline\n"
+                     "• `ken burns` — video animation technique\n"
+                     "• `hf models` — AI model chain\n"
+                     "• `unique features` — what makes this special\n"
+                     "• `error` — troubleshooting\n"
+                     "• `template` — save/load settings\n"
+                     "• `audio` — music & voice system\n"
+                     "• `styles` — Premium/Energetic/Fun")
+    
+    history = history or []
+    history.append({"role": "user", "content": user_msg})
+    history.append({"role": "assistant", "content": reply})
+    return history, ""
 
 # ══════════════════════════════════════════════════════════════════
 #  HF VIDEO CHAIN
@@ -210,7 +307,6 @@ def get_video(pil, prompt, dur, cb=None):
         time.sleep(0.5)
     return ken_burns(pil, duration_sec=dur), "Ken Burns"
 
-
 # ══════════════════════════════════════════════════════════════════
 #  KEN BURNS
 # ══════════════════════════════════════════════════════════════════
@@ -230,7 +326,6 @@ def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
     img_r=ImageEnhance.Contrast(img_r).enhance(1.05)
     img_r=ImageEnhance.Color(img_r).enhance(1.08)
 
-    # Blurred bg
     bg=img.resize((TW,TH),Image.LANCZOS).filter(ImageFilter.GaussianBlur(18))
     bg=ImageEnhance.Brightness(bg).enhance(0.55)
     canvas=bg.copy(); canvas.paste(img_r,((TW-nw)//2,(TH-nh)//2))
@@ -287,21 +382,15 @@ def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
     writer.release()
     return tmp.name
 
-
 # ══════════════════════════════════════════════════════════════════
 #  MULTI-VIDEO MERGE
 # ══════════════════════════════════════════════════════════════════
 def merge_videos(paths):
-    """Concatenate multiple mp4s with crossfade using ffmpeg."""
     if len(paths)==1: return paths[0]
     out=paths[0].replace(".mp4","_merged.mp4")
-
-    # Write concat list
     lst=tempfile.NamedTemporaryFile(suffix=".txt",mode="w",delete=False)
     for p in paths: lst.write(f"file '{p}'\n")
     lst.flush()
-
-    # Simple concat (re-encode for compatibility)
     ret=os.system(
         f'ffmpeg -y -f concat -safe 0 -i "{lst.name}" '
         f'-c:v libx264 -c:a aac -b:a 128k -movflags +faststart '
@@ -309,9 +398,8 @@ def merge_videos(paths):
     )
     return out if (ret==0 and os.path.exists(out)) else paths[-1]
 
-
 # ══════════════════════════════════════════════════════════════════
-#  CAPTIONS  (ffmpeg drawtext)
+#  CAPTIONS
 # ══════════════════════════════════════════════════════════════════
 def add_captions_ffmpeg(video_path, caption, duration_sec, style):
     def clean(t): return re.sub(r"[^A-Za-z0-9 !.,\-\u0900-\u097F]","",t).strip()
@@ -342,7 +430,6 @@ def add_captions_ffmpeg(video_path, caption, duration_sec, style):
     ret=os.system(f'ffmpeg -y -i "{video_path}" -vf "{vf}" -c:a copy "{out}" -loglevel error')
     return out if (ret==0 and os.path.exists(out)) else video_path
 
-
 # ══════════════════════════════════════════════════════════════════
 #  AUDIO
 # ══════════════════════════════════════════════════════════════════
@@ -390,27 +477,49 @@ def add_audio(video_path, caption, duration_sec, style):
     os.system(f'ffmpeg -y -i "{video_path}" -i "{audio}" -c:v copy -c:a aac -b:a 128k -shortest "{final}" -loglevel error')
     return final if os.path.exists(final) else video_path
 
-
 # ══════════════════════════════════════════════════════════════════
-#  MAIN PIPELINE
+#  MAIN PIPELINE  (FIXED: safe image conversion)
 # ══════════════════════════════════════════════════════════════════
 def generate(images, caption, style, language, duration, add_aud, add_cap, progress=gr.Progress()):
-    # Filter out None images
-    pils = [img if isinstance(img,Image.Image) else Image.fromarray(img)
-            for img in (images or []) if img is not None]
-    if not pils: return None, "⚠️ Upload at least 1 image!", "No image provided."
+    # ✅ FIX: Safe multi-format image handling
+    pils = []
+    if images:
+        for img in images:
+            if img is None:
+                continue
+            try:
+                if isinstance(img, Image.Image):
+                    pils.append(img.convert("RGB"))
+                elif isinstance(img, np.ndarray):
+                    pils.append(Image.fromarray(img).convert("RGB"))
+                elif isinstance(img, dict):
+                    # Gradio sometimes wraps as dict
+                    raw = img.get("composite") or img.get("image") or img.get("path")
+                    if raw is not None:
+                        if isinstance(raw, np.ndarray):
+                            pils.append(Image.fromarray(raw).convert("RGB"))
+                        elif isinstance(raw, Image.Image):
+                            pils.append(raw.convert("RGB"))
+                        elif isinstance(raw, str) and os.path.exists(raw):
+                            pils.append(Image.open(raw).convert("RGB"))
+                elif isinstance(img, str) and os.path.exists(img):
+                    pils.append(Image.open(img).convert("RGB"))
+            except Exception as e:
+                print(f"  ⚠️ Skipping image: {e}")
+                continue
+
+    if not pils:
+        return None, "⚠️ Upload at least 1 valid image!", "No image provided."
 
     cap = caption.strip() or ""
     dur = int(duration)
     lines = []
-    def log(msg): lines.append(msg); progress(min(.05+len(lines)*.08,.80),desc=msg)
+    def log(msg): lines.append(msg); progress(min(.05+len(lines)*.08,.80), desc=msg)
 
-    # ── Auto-detect from FIRST image ──────────────────────────────
     progress(.02, desc="🔍 Auto-detecting category...")
     category, auto_prompt, detected_label = auto_detect(pils[0], cap)
     log(f"🔍 Detected: {detected_label or category}")
 
-    # If caption empty, auto-generate one
     if not cap:
         cap_hints = {
             "Fashion":"Step into style. Own the moment.",
@@ -424,19 +533,15 @@ def generate(images, caption, style, language, duration, add_aud, add_cap, progr
         cap = cap_hints.get(category,"Premium quality. Shop now.")
         log(f"💡 Auto caption: {cap}")
 
-    # ── Get insights ───────────────────────────────────────────────
     insight, ai_cap = get_insights(category, style, language, cap)
 
-    # ── Generate video per image ───────────────────────────────────
     video_paths = []
-    clip_dur = max(4, dur // len(pils))   # split duration across images
+    clip_dur = max(4, dur // len(pils))
 
     for idx, pil in enumerate(pils):
         log(f"🎬 Image {idx+1}/{len(pils)}...")
-        # Re-detect for each image but use same prompt style
         _, img_prompt, _ = auto_detect(pil, cap)
         full_prompt = f"{img_prompt}, {cap[:60]}"
-
         vpath, model = get_video(pil, full_prompt, clip_dur, cb=log if idx==0 else None)
 
         if add_cap:
@@ -447,14 +552,12 @@ def generate(images, caption, style, language, duration, add_aud, add_cap, progr
         video_paths.append(vpath)
         log(f"✅ Clip {idx+1} done ({model})")
 
-    # ── Merge if multiple ─────────────────────────────────────────
     if len(video_paths) > 1:
         log("🔗 Merging clips...")
         final = merge_videos(video_paths)
     else:
         final = video_paths[0]
 
-    # ── Audio on merged video ─────────────────────────────────────
     if add_aud:
         log("🎵 Adding music + voice...")
         final = add_audio(final, cap, dur, style.lower())
@@ -462,64 +565,255 @@ def generate(images, caption, style, language, duration, add_aud, add_cap, progr
     progress(1.0, desc="✅ Done!")
     return final, "\n".join(lines), insight
 
-
 # ══════════════════════════════════════════════════════════════════
 #  UI
 # ══════════════════════════════════════════════════════════════════
-css="""
-#title{text-align:center;font-size:2.3rem;font-weight:900}
-#sub{text-align:center;color:#999;margin-bottom:1.2rem;font-size:1rem}
+css = """
+#title{text-align:center;font-size:2.3rem;font-weight:900;background:linear-gradient(135deg,#a855f7,#ec4899);-webkit-background-clip:text;-webkit-text-fill-color:transparent}
+#sub{text-align:center;color:#aaa;margin-bottom:1.2rem;font-size:1rem}
 .insight{font-family:monospace;font-size:.86rem;line-height:1.75}
+.bot-container{border:1px solid #3a3a5c;border-radius:12px;padding:0;overflow:hidden}
+.save-row{gap:8px}
+.feature-badge{display:inline-block;background:linear-gradient(135deg,#7c3aed,#db2777);color:white;padding:2px 10px;border-radius:99px;font-size:.75rem;margin:2px}
+.tab-label{font-weight:700}
 """
+
 with gr.Blocks(css=css, theme=gr.themes.Soft(primary_hue="violet")) as demo:
     gr.Markdown("# 🎬 AI Reel Generator", elem_id="title")
-    gr.Markdown("Upload 1-5 images → AI auto-detects category → cinematic reel + smart posting strategy", elem_id="sub")
-
-    with gr.Row():
-        # ── LEFT ──────────────────────────────────────────────────
-        with gr.Column(scale=1):
-            img_in = gr.Gallery(
-                label="📸 Upload 1–5 Images (drag & drop)",
-                type="pil",
-                columns=5, rows=1,
-                height=200,
-                object_fit="contain",
-            )
-            cap_in = gr.Textbox(
-                label="✏️ Caption / Description (leave blank = auto-detect)",
-                placeholder="e.g. Premium sneakers with star design... or leave empty!",
-                lines=2,
-            )
-            with gr.Row():
-                sty_dd  = gr.Dropdown(["Premium","Energetic","Fun"], value="Premium", label="🎨 Style")
-                lang_dd = gr.Dropdown(["English","Hindi","Hinglish"], value="English",  label="🌐 Language")
+    gr.Markdown(
+        "Upload 1-5 images → AI auto-detects category → cinematic reel + smart posting strategy\n\n"
+        '<span class="feature-badge">Multi-Image</span>'
+        '<span class="feature-badge">Multilingual</span>'
+        '<span class="feature-badge">AI Chain</span>'
+        '<span class="feature-badge">Template Save/Share</span>'
+        '<span class="feature-badge">ReelBot 🤖</span>',
+        elem_id="sub"
+    )
 
-            dur_sl = gr.Slider(minimum=5, maximum=20, value=6, step=1,
-                               label="⏱️ Total Duration (seconds)")
+    with gr.Tabs():
+        # ── TAB 1: GENERATOR ─────────────────────────────────────
+        with gr.Tab("🎬 Generator", elem_classes="tab-label"):
             with gr.Row():
-                aud_cb = gr.Checkbox(label="🎵 Music + Voice", value=True)
-                cap_cb = gr.Checkbox(label="💬 Captions",      value=True)
+                # LEFT
+                with gr.Column(scale=1):
+                    img_in = gr.Gallery(
+                        label="📸 Upload 1–5 Images (drag & drop)",
+                        type="pil",
+                        columns=5, rows=1,
+                        height=200,
+                        object_fit="contain",
+                    )
+                    cap_in = gr.Textbox(
+                        label="✏️ Caption / Description (leave blank = auto-detect)",
+                        placeholder="e.g. Premium sneakers with star design... or leave empty!",
+                        lines=2,
+                    )
+                    with gr.Row():
+                        sty_dd  = gr.Dropdown(["Premium","Energetic","Fun"], value="Premium", label="🎨 Style")
+                        lang_dd = gr.Dropdown(["English","Hindi","Hinglish"], value="English", label="🌐 Language")
+
+                    dur_sl = gr.Slider(minimum=5, maximum=20, value=6, step=1,
+                                       label="⏱️ Total Duration (seconds)")
+                    with gr.Row():
+                        aud_cb = gr.Checkbox(label="🎵 Music + Voice", value=True)
+                        cap_cb = gr.Checkbox(label="💬 Captions",      value=True)
+
+                    gen_btn = gr.Button("🚀 Generate Reel + Smart Insights", variant="primary", size="lg")
+                    gr.Markdown(
+                        "**🔗 AI Chain:** LTX-2 ⚡ → Wan 2.2 → SVD-XT → Kling → LTX-Video → Ken Burns ✅\n\n"
+                        "💡 Upload multiple images for a multi-clip reel!"
+                    )
+
+                # RIGHT
+                with gr.Column(scale=1):
+                    vid_out     = gr.Video(label="🎥 Cinematic Reel", height=400)
+                    insight_out = gr.Textbox(
+                        label="📊 Smart Insights",
+                        lines=16, interactive=False, elem_classes="insight",
+                    )
+                    log_out = gr.Textbox(label="🔧 Log", lines=4, interactive=False)
+
+        # ── TAB 2: TEMPLATES ─────────────────────────────────────
+        with gr.Tab("💾 Templates", elem_classes="tab-label"):
+            gr.Markdown("### 💾 Save, Load & Share Your Reel Settings")
+
+            with gr.Row(elem_classes="save-row"):
+                tpl_name_in = gr.Textbox(label="Template Name", placeholder="e.g. My Brand Style", scale=3)
+                save_btn    = gr.Button("💾 Save Current Settings", variant="primary", scale=1)
+
+            tpl_status   = gr.Textbox(label="Status", interactive=False, lines=1)
+            tpl_list     = gr.Dropdown(label="📂 Saved Templates", choices=get_template_names(), interactive=True)
 
-            gen_btn = gr.Button("🚀 Generate Reel + Smart Insights", variant="primary", size="lg")
-            gr.Markdown(
-                "**🔗 AI Chain:** LTX-2 ⚡ → Wan 2.2 → SVD-XT → Kling → LTX-Video → Ken Burns ✅\n\n"
-                "💡 Upload multiple images for a multi-clip reel!"
+            with gr.Row():
+                load_btn   = gr.Button("📂 Load Template", variant="secondary")
+                del_btn    = gr.Button("🗑️ Delete Template", variant="stop")
+                export_btn = gr.Button("📤 Export as JSON")
+
+            export_file = gr.File(label="⬇️ Download Template JSON", visible=True)
+
+            gr.Markdown("""
+            **How to use Templates:**
+            1. Configure your settings in the Generator tab
+            2. Give it a name and click **Save Current Settings**
+            3. Next time, just pick from the dropdown and **Load Template**
+            4. **Export** to share with teammates or save as backup
+            """)
+
+        # ── TAB 3: REELBOT ───────────────────────────────────────
+        with gr.Tab("🤖 ReelBot", elem_classes="tab-label"):
+            gr.Markdown("""
+            ### 🤖 ReelBot — Your AI Project Guide
+            _Ask me anything about how this project works, the tech used, features, and more!_
+            """)
+
+            bot_chatbox = gr.Chatbot(
+                label="💬 Chat with ReelBot",
+                height=420,
+                type="messages",
+                avatar_images=(None, "https://api.dicebear.com/7.x/bottts/svg?seed=reelbot"),
+                value=[{
+                    "role": "assistant",
+                    "content": (
+                        "👋 **Namaste! Main hun ReelBot!** 🤖\n\n"
+                        "Main is project ke baare mein sab kuch jaanta hun.\n\n"
+                        "**Mujhse poocho:**\n"
+                        "• `how does this work` — Full pipeline samjho\n"
+                        "• `ken burns` — Animation technique\n"
+                        "• `hf models` — AI model chain\n"
+                        "• `unique features` — Kya khaas hai is project mein\n"
+                        "• `error` — Bug troubleshooting\n"
+                        "• `template` — Settings save/share\n"
+                        "• `audio` — Music generation\n"
+                        "• `styles` — Premium/Energetic/Fun\n\n"
+                        "**Koi bhi sawaal pucho! 🚀**"
+                    )
+                }]
             )
 
-        # ── RIGHT ─────────────────────────────────────────────────
-        with gr.Column(scale=1):
-            vid_out     = gr.Video(label="🎥 Cinematic Reel", height=400)
-            insight_out = gr.Textbox(
-                label="📊 Smart Insights — Auto-Detected + Audience + Posting Time + AI Caption",
-                lines=16, interactive=False, elem_classes="insight",
-            )
-            log_out = gr.Textbox(label="🔧 Log", lines=4, interactive=False)
+            with gr.Row():
+                bot_input = gr.Textbox(
+                    placeholder="Ask: 'how does this work?' or 'ken burns kya hai?' or 'unique features kya hain?'",
+                    label="Your Question",
+                    scale=5,
+                )
+                bot_send = gr.Button("Send 📨", variant="primary", scale=1)
 
+            with gr.Row():
+                gr.Button("how does this work").click(
+                    lambda h: bot_reply("how does this work", h),
+                    inputs=[bot_chatbox], outputs=[bot_chatbox, bot_input]
+                )
+                gr.Button("ken burns kya hai").click(
+                    lambda h: bot_reply("ken burns", h),
+                    inputs=[bot_chatbox], outputs=[bot_chatbox, bot_input]
+                )
+                gr.Button("unique features").click(
+                    lambda h: bot_reply("unique features", h),
+                    inputs=[bot_chatbox], outputs=[bot_chatbox, bot_input]
+                )
+                gr.Button("error fix").click(
+                    lambda h: bot_reply("error fix", h),
+                    inputs=[bot_chatbox], outputs=[bot_chatbox, bot_input]
+                )
+
+        # ── TAB 4: TECH EXPLAINED ────────────────────────────────
+        with gr.Tab("📚 Tech Stack", elem_classes="tab-label"):
+            gr.Markdown("""
+            ## 🛠️ Technology Used — Full Breakdown
+
+            ### 🎬 Video Generation
+            | Component | Technology | Purpose |
+            |-----------|-----------|---------|
+            | **Ken Burns Effect** | OpenCV + NumPy | Cinematic zoom/pan animation |
+            | **Color Grading** | NumPy array ops | Style-based color correction |
+            | **Vignette** | NumPy distance map | Cinematic edge darkening |
+            | **Video Encoding** | OpenCV VideoWriter | MP4 output @ 30fps |
+            | **AI Video** | HuggingFace InferenceClient | Image-to-video (when available) |
+
+            ### 🤗 AI Model Chain
+            | Priority | Model | Provider | Type |
+            |----------|-------|----------|------|
+            | 1 | LTX-2 ⚡ | Lightricks | Fast I2V |
+            | 2 | Wan 2.2 | Wan-AI | High quality I2V |
+            | 3 | SVD-XT | Stability AI | Stable Video Diffusion |
+            | 4 | Kling | KlingTeam | LivePortrait |
+            | 5 | LTX-Video | Lightricks | Fallback I2V |
+            | 6 ✅ | Ken Burns | Local | Always works! |
+
+            ### 🎵 Audio System
+            | Component | Technology | Details |
+            |-----------|-----------|---------|
+            | **BGM Generation** | NumPy + wave | Sine waves, kick drum, hi-hat |
+            | **TTS Voice** | gTTS (Google TTS) | Caption narration |
+            | **Audio Mixing** | ffmpeg amix | BGM 20% + Voice 95% |
+            | **BPM by Style** | Custom logic | Premium=88, Energetic=126, Fun=104 |
+
+            ### 💬 Caption System
+            | Feature | Technology |
+            |---------|-----------|
+            | Text Overlay | ffmpeg drawtext filter |
+            | Fade Animation | ffmpeg alpha expression |
+            | Font | DejaVu / Liberation Sans Bold |
+            | Languages | English / Hindi / Hinglish |
+
+            ### 🔍 Auto-Detection
+            | Step | Technology |
+            |------|-----------|
+            | Image Classification | google/vit-base-patch16-224 |
+            | Label Mapping | Custom Python dict |
+            | Caption Fallback | Keyword matching |
+
+            ### 🌟 Unique Points
+            > ✅ **No GPU required** — Ken Burns always as fallback
+            > ✅ **Multilingual** — Hindi captions with Devanagari support
+            > ✅ **Programmatic BGM** — No audio files needed
+            > ✅ **Template system** — Save/load/export settings as JSON
+            > ✅ **AI fallback chain** — 5 models tried before local fallback
+            > ✅ **ReelBot** — Built-in explainer chatbot
+            > ✅ **Multi-image merge** — Up to 5 clips concatenated
+            > ✅ **Auto posting strategy** — AI-driven best time recommendation
+            """)
+
+    # ── EVENTS ────────────────────────────────────────────────────
     gen_btn.click(
         fn=generate,
         inputs=[img_in, cap_in, sty_dd, lang_dd, dur_sl, aud_cb, cap_cb],
         outputs=[vid_out, log_out, insight_out],
     )
 
+    # Template events
+    save_btn.click(
+        fn=save_template,
+        inputs=[tpl_name_in, sty_dd, lang_dd, dur_sl, cap_in, aud_cb, cap_cb],
+        outputs=[tpl_status, tpl_list],
+    )
+    load_btn.click(
+        fn=load_template,
+        inputs=[tpl_list],
+        outputs=[sty_dd, lang_dd, dur_sl, cap_in, aud_cb, cap_cb],
+    )
+    del_btn.click(
+        fn=delete_template,
+        inputs=[tpl_list],
+        outputs=[tpl_status, tpl_list],
+    )
+    export_btn.click(
+        fn=export_template,
+        inputs=[tpl_list],
+        outputs=[export_file],
+    )
+
+    # Bot events
+    bot_send.click(
+        fn=bot_reply,
+        inputs=[bot_input, bot_chatbox],
+        outputs=[bot_chatbox, bot_input],
+    )
+    bot_input.submit(
+        fn=bot_reply,
+        inputs=[bot_input, bot_chatbox],
+        outputs=[bot_chatbox, bot_input],
+    )
+
 if __name__ == "__main__":
     demo.launch()