Update app.py

app.py

@@ -1,4 +1,4 @@
-import os, tempfile, io, math, time, threading
+import os, tempfile, io, math, time, threading, re, random
 import numpy as np
 import cv2
 import gradio as gr
@@ -14,100 +14,236 @@ if hf_token:
         print("✅ HF ready")
     except Exception as e: print(f"⚠️ HF: {e}")
 
-# ── HF MODELS ─────────────────────────────────────────────────────
+# ══════════════════════════════════════════════════════════════════
+#  AUTO-DETECT from image (color + HF classifier)
+# ══════════════════════════════════════════════════════════════════
+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",
+            )
+            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",
+    }
+    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"
+        elif any(w in cap_low for w in ["food","eat","recipe","cook","restaurant","cafe","pizza","biryani"]): category="Food"
+        elif any(w in cap_low for w in ["phone","laptop","tech","gadget","device","app","camera"]): category="Tech"
+        elif any(w in cap_low for w in ["skin","beauty","makeup","lipstick","cream","hair","glow"]): category="Beauty"
+        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",
+        "Tech":      "cinematic tech product reveal, sleek background, blue accent lighting, smooth rotation, premium feel",
+        "Beauty":    "cinematic beauty product shot, soft pink bokeh, gentle sparkle, slow zoom, elegant lighting",
+        "Fitness":   "cinematic fitness product shot, energetic motion blur, bold lighting, dynamic angle, powerful",
+        "Lifestyle": "cinematic lifestyle shot, warm ambient light, cozy aesthetic, slow pan, aspirational feel",
+        "Product/Other": "cinematic product advertisement, dramatic lighting, smooth zoom, professional commercial look",
+    }
+    auto_prompt = prompts.get(category, prompts["Product/Other"])
+    if label: auto_prompt = f"{label} product, {auto_prompt}"
+
+    return category, auto_prompt, label
+
+
+# ══════════════════════════════════════════════════════════════════
+#  SMART INSIGHTS
+# ════════════════════════════════════════════════��═════════════════
+POSTING_TIMES = {
+    "Fashion":      {"best":"9:00 PM",  "days":"Tue, Thu, Fri", "slots":["7AM","12PM","6PM","9PM"]},
+    "Food":         {"best":"12:00 PM", "days":"Mon, Wed, Sat", "slots":["11AM","1PM","7PM"]},
+    "Tech":         {"best":"8:00 AM",  "days":"Mon, Tue, Wed", "slots":["8AM","12PM","5PM"]},
+    "Beauty":       {"best":"8:00 PM",  "days":"Wed, Fri, Sun", "slots":["8AM","1PM","8PM"]},
+    "Fitness":      {"best":"6:00 AM",  "days":"Mon, Wed, Fri", "slots":["6AM","12PM","7PM"]},
+    "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",
+    "Tech":         "💻 20-40 · Tech enthusiasts · Early adopters · Gadget buyers",
+    "Beauty":       "💄 16-35 yo · Beauty lovers · Skincare fans · Self-care community",
+    "Fitness":      "💪 18-40 · Gym goers · Health-conscious · Athletes · Wellness seekers",
+    "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",
+                      "Elevate your game. {cap} 💫 DM to order!"],
+        "Energetic": ["🔥 {cap} Hit different. Grab yours NOW 👆 Limited stock!",
+                      "⚡ Game changer! {cap} Don't sleep on this 🚀"],
+        "Fun":       ["Obsessed!! 😍 {cap} Tag someone who needs this 👇",
+                      "POV: You just found your new fav 🎉 {cap} Link in bio!"],
+    },
+    "Hindi": {
+        "Premium":   ["✨ {cap} क्वालिटी जो बोलती है। 🛒 अभी खरीदें → Bio में link",
+                      "अपना स्टाइल बढ़ाएं। {cap} 💫 Order के लिए DM करें!"],
+        "Energetic": ["🔥 {cap} एकदम अलग! अभी grab करो 👆 Limited stock!",
+                      "⚡ Game changer! {cap} मत सोचो, order करो 🚀"],
+        "Fun":       ["दीवाने हो जाओगे!! 😍 {cap} किसी को tag करो 👇",
+                      "POV: नया favourite मिल गया 🎉 {cap} Bio में link!"],
+    },
+    "Hinglish": {
+        "Premium":   ["✨ {cap} Quality toh dekho yaar! 🛒 Shop karo → Bio mein link",
+                      "Style upgrade time! {cap} 💫 DM karo order ke liye!"],
+        "Energetic": ["🔥 {cap} Bilkul alag hai bhai! Abhi lo 👆 Limited stock!",
+                      "⚡ Ek dum fire! {cap} Mat ruko, order karo 🚀"],
+        "Fun":       ["Yaar yeh toh kamaal hai!! 😍 {cap} Kisi ko tag karo 👇",
+                      "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",
+    "Tech":      "#TechReview #Gadgets #TechLovers #Innovation #NewTech #MustHave",
+    "Beauty":    "#BeautyTips #Skincare #MakeupLovers #GlowUp #BeautyReels #GRWM",
+    "Fitness":   "#FitnessMotivation #GymLife #HealthyLifestyle #FitFam #WorkoutReels",
+    "Lifestyle": "#Lifestyle #Aesthetic #HomeDecor #VibeCheck #DailyInspo #Reels",
+    "Product/Other":"#NewProduct #MustHave #ShopNow #Trending #Viral #Reels #ForYou",
+}
+
+def get_insights(category, style, language, cap):
+    pt = POSTING_TIMES[category]
+    clean_cap = re.sub(r"[^\w\s!.,'-]","",cap).strip()[:60]
+    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"
+        f"🎯 Category: {category}\n\n"
+        f"👥 Target Audience:\n{AUDIENCES[category]}\n\n"
+        f"⏰ Best Time to Post:\n"
+        f"🏆 Prime: {pt['best']}  |  📅 Days: {pt['days']}\n"
+        f"🕐 All slots: {', '.join(pt['slots'])}\n\n"
+        f"💬 AI Caption ({language} · {style}):\n{ai_cap}\n\n"
+        f"#️⃣  Hashtags:\n{tags}\n"
+        f"{'━'*38}"
+    )
+    return insight, ai_cap
+
+
+# ══════════════════════════════════════════════════════════════════
+#  HF VIDEO CHAIN
+# ══════════════════════════════════════════════════════════════════
 HF_MODELS = [
-    {"id": "Lightricks/LTX-2",                                 "name": "LTX-2 ⚡"},
-    {"id": "Wan-AI/Wan2.2-I2V-A14B",                           "name": "Wan 2.2"},
-    {"id": "stabilityai/stable-video-diffusion-img2vid-xt",    "name": "SVD-XT"},
-    {"id": "KlingTeam/LivePortrait",                           "name": "Kling LivePortrait"},
-    {"id": "Lightricks/LTX-Video",                             "name": "LTX-Video"},
-    {"id": "__local__",                                        "name": "Ken Burns ✅"},
+    {"id":"Lightricks/LTX-2",                              "name":"LTX-2 ⚡"},
+    {"id":"Wan-AI/Wan2.2-I2V-A14B",                        "name":"Wan 2.2"},
+    {"id":"stabilityai/stable-video-diffusion-img2vid-xt", "name":"SVD-XT"},
+    {"id":"KlingTeam/LivePortrait",                        "name":"Kling"},
+    {"id":"Lightricks/LTX-Video",                          "name":"LTX-Video"},
+    {"id":"__local__",                                     "name":"Ken Burns ✅"},
 ]
 
-def pil_to_bytes(img):
-    b=io.BytesIO(); img.save(b,format="JPEG",quality=92); return b.getvalue()
-
 def run_timeout(fn, sec, *a, **kw):
     box=[None]; err=[None]
     def r():
         try: box[0]=fn(*a,**kw)
         except Exception as e: err[0]=str(e)
     t=threading.Thread(target=r,daemon=True); t.start(); t.join(timeout=sec)
-    if t.is_alive(): print(f"  ⏱ timeout"); return None
-    if err[0]: print(f"  ❌ {err[0][:80]}")
+    if t.is_alive(): return None
     return box[0]
 
 def try_hf(model_id, pil, prompt):
     if not hf_client: return None
     try:
-        r=hf_client.image_to_video(image=pil_to_bytes(pil),model=model_id,prompt=prompt)
+        b=io.BytesIO(); pil.save(b,format="JPEG",quality=92)
+        r=hf_client.image_to_video(image=b.getvalue(),model=model_id,prompt=prompt)
         return r.read() if hasattr(r,"read") else r
     except Exception as e: print(f"  ❌ {model_id}: {e}"); return None
 
-def get_video(pil, prompt, cb=None):
+def get_video(pil, prompt, dur, cb=None):
     for m in HF_MODELS:
         mid,mname=m["id"],m["name"]
         if cb: cb(f"⏳ Trying: {mname}")
         if mid=="__local__":
-            return ken_burns(pil), mname
+            return ken_burns(pil, duration_sec=dur), mname
         data=run_timeout(try_hf,50,mid,pil,prompt)
         if data:
             t=tempfile.NamedTemporaryFile(suffix=".mp4",delete=False)
             t.write(data); t.flush()
             return t.name, mname
-        time.sleep(1)
-    return ken_burns(pil), "Ken Burns"
+        time.sleep(0.5)
+    return ken_burns(pil, duration_sec=dur), "Ken Burns"
 
 
 # ══════════════════════════════════════════════════════════════════
-#  KEN BURNS  (working, image always shows)
+#  KEN BURNS
 # ══════════════════════════════════════════════════════════════════
-def ease(t): t=max(0.,min(1.,t)); return t*t*(3-2*t)
-def ease_cubic(t): t=max(0.,min(1.,t)); return 4*t*t*t if t<.5 else 1-math.pow(-2*t+2,3)/2
-def ease_expo(t): return 1-math.pow(2,-10*t) if t<1 else 1.
-def ease_bounce(t):
-    if t<1/2.75: return 7.5625*t*t
-    elif t<2/2.75: t-=1.5/2.75; return 7.5625*t*t+.75
-    elif t<2.5/2.75: t-=2.25/2.75; return 7.5625*t*t+.9375
-    else: t-=2.625/2.75; return 7.5625*t*t+.984375
+def ease_c(t): t=max(0.,min(1.,t)); return 4*t*t*t if t<.5 else 1-math.pow(-2*t+2,3)/2
+def ease_e(t): return 1-math.pow(2,-10*t) if t<1 else 1.
+def ease_s(t): t=max(0.,min(1.,t)); return t*t*(3-2*t)
 
 def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
-    TW,TH=720,1280
-    # Small pad — just enough for gentle movement, no aggressive zoom
-    pad=60; BW,BH=TW+pad*2,TH+pad*2
-    total=duration_sec*fps
+    TW,TH=720,1280; pad=60; BW,BH=TW+pad*2,TH+pad*2
+    total=int(duration_sec*fps)
 
-    # Prepare image — fit full image, letterbox if needed
     img=pil.convert("RGB"); sw,sh=img.size
-    # Fit entire image inside TH height, pad sides with blurred bg
-    scale=TH/sh; nw=int(sw*scale); nh=TH
-    if nw>TW: scale=TW/sw; nw=TW; nh=int(sh*scale)
-    img_resized=img.resize((nw,nh),Image.LANCZOS)
-    # Blurred background fill
-    bg=img.resize((TW,TH),Image.LANCZOS)
-    bg=bg.filter(ImageFilter.GaussianBlur(radius=20))
-    bg_arr=np.array(ImageEnhance.Brightness(bg).enhance(0.5))
-    canvas=Image.fromarray(bg_arr)
-    # Paste sharp image centered
-    px=(TW-nw)//2; py=(TH-nh)//2
-    canvas.paste(img_resized,(px,py))
-    canvas=canvas.filter(ImageFilter.UnsharpMask(radius=0.8,percent=110,threshold=2))
-    canvas=ImageEnhance.Contrast(canvas).enhance(1.05)
-    canvas=ImageEnhance.Color(canvas).enhance(1.08)
+    scale=min(TH/sh, TW/sw)
+    nw,nh=int(sw*scale),int(sh*scale)
+    img_r=img.resize((nw,nh),Image.LANCZOS)
+    img_r=img_r.filter(ImageFilter.UnsharpMask(radius=0.8,percent=110,threshold=2))
+    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))
     base=np.array(canvas.resize((BW,BH),Image.LANCZOS))
 
-    # Pre-baked vignette mask (very subtle)
     Y,X=np.ogrid[:TH,:TW]
     dist=np.sqrt(((X-TW/2)/(TW/2))**2+((Y-TH/2)/(TH/2))**2)
     vmask=np.clip(1.-0.22*np.maximum(dist-0.85,0)**2,0,1).astype(np.float32)
 
-    # GENTLE zoom: 1.00→1.06 max — full image always visible
-    SEG=[
-        (0.00,0.30, 1.00,1.04,  0,          -int(pad*.40),  0,          -int(pad*.40)),
-        (0.30,0.60, 1.04,1.06, -int(pad*.30), int(pad*.30), -int(pad*.40),-int(pad*.70)),
-        (0.60,0.80, 1.06,1.04,  int(pad*.30), int(pad*.50), -int(pad*.70),-int(pad*.40)),
-        (0.80,1.00, 1.04,1.00,  int(pad*.50), 0,            -int(pad*.40), 0),
-    ]
+    SEG=[(0.00,0.30,1.00,1.04,0,-int(pad*.4),0,-int(pad*.4)),
+         (0.30,0.60,1.04,1.06,-int(pad*.3),int(pad*.3),-int(pad*.4),-int(pad*.7)),
+         (0.60,0.80,1.06,1.04,int(pad*.3),int(pad*.5),-int(pad*.7),-int(pad*.4)),
+         (0.80,1.00,1.04,1.00,int(pad*.5),0,-int(pad*.4),0)]
 
     tmp=tempfile.NamedTemporaryFile(suffix=".mp4",delete=False)
     writer=cv2.VideoWriter(tmp.name,cv2.VideoWriter_fourcc(*"mp4v"),fps,(TW,TH))
@@ -117,10 +253,9 @@ def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
         zoom=pan_x=pan_y=None
         for t0,t1,z0,z1,px0,px1,py0,py1 in SEG:
             if t0<=tg<=t1:
-                te=ease_cubic((tg-t0)/(t1-t0))
+                te=ease_c((tg-t0)/(t1-t0))
                 zoom=z0+(z1-z0)*te; pan_x=int(px0+(px1-px0)*te); pan_y=int(py0+(py1-py0)*te); break
         if zoom is None: zoom,pan_x,pan_y=1.,0,0
-        # No shake — keeps image stable and well-framed
 
         cw,ch=int(TW/zoom),int(TH/zoom)
         ox,oy=BW//2+pan_x,BH//2+pan_y
@@ -130,31 +265,21 @@ def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
 
         frame=cv2.resize(base[y1:y2,x1:x2],(TW,TH),interpolation=cv2.INTER_LINEAR)
 
-        # Very subtle color grade
         f=frame.astype(np.float32)/255.
         if style=="premium":
-            f[:,:,0]=np.clip(f[:,:,0]*1.03+.01,0,1)
-            f[:,:,2]=np.clip(f[:,:,2]*1.02,0,1)
+            f[:,:,0]=np.clip(f[:,:,0]*1.03+.01,0,1); f[:,:,2]=np.clip(f[:,:,2]*1.02,0,1)
         elif style=="energetic":
-            gray=0.299*f[:,:,0:1]+0.587*f[:,:,1:2]+0.114*f[:,:,2:3]
-            f=np.clip(gray+1.2*(f-gray),0,1); f=np.clip(f*1.04,0,1)
+            g=0.299*f[:,:,0:1]+0.587*f[:,:,1:2]+0.114*f[:,:,2:3]
+            f=np.clip(g+1.2*(f-g),0,1); f=np.clip(f*1.04,0,1)
         elif style=="fun":
-            f[:,:,0]=np.clip(f[:,:,0]*1.05,0,1)
-            f[:,:,1]=np.clip(f[:,:,1]*1.03,0,1)
+            f[:,:,0]=np.clip(f[:,:,0]*1.05,0,1); f[:,:,1]=np.clip(f[:,:,1]*1.03,0,1)
         frame=np.clip(f*255,0,255).astype(np.uint8)
-
-        # Vignette
         frame=np.clip(frame.astype(np.float32)*vmask[:,:,None],0,255).astype(np.uint8)
-
-        # Grain
-        frame=np.clip(frame.astype(np.float32)+np.random.normal(0,3,frame.shape),0,255).astype(np.uint8)
-
-        # Bars
+        frame=np.clip(frame.astype(np.float32)+np.random.normal(0,2.5,frame.shape),0,255).astype(np.uint8)
         frame[:36,:]=0; frame[-36:,:]=0
 
-        # Fade in (2%) / out (5%)
-        if tg<0.02: alpha=ease_expo(tg/0.02)
-        elif tg>0.95: alpha=ease(1-(tg-0.95)/0.05)
+        if tg<0.02: alpha=ease_e(tg/0.02)
+        elif tg>0.95: alpha=ease_s(1-(tg-0.95)/0.05)
         else: alpha=1.
         if alpha<1.: frame=np.clip(frame.astype(np.float32)*alpha,0,255).astype(np.uint8)
 
@@ -164,100 +289,86 @@ def ken_burns(pil, duration_sec=6, fps=30, style="premium"):
 
 
 # ══════════════════════════════════════════════════════════════════
-#  CAPTIONS — burn into existing video via ffmpeg
+#  MULTI-VIDEO MERGE
 # ══════════════════════════════════════════════════════════════════
-def add_captions_ffmpeg(video_path, caption, duration_sec, style):
-    """Burn animated captions + hashtag tag + shop-now CTA using ffmpeg drawtext."""
-    import re
-    def clean(t): return re.sub(r"[^A-Za-z0-9 !.,-]","",t).strip()
+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 '
+        f'"{out}" -loglevel error'
+    )
+    return out if (ret==0 and os.path.exists(out)) else paths[-1]
 
-    words=caption.strip().split()
-    mid=max(1,len(words)//2)
-    line1=clean(" ".join(words[:mid]))
-    line2=clean(" ".join(words[mid:])) if len(words)>1 else line1
 
-    colors={"premium":"FFD232","energetic":"3CC8FF","fun":"FF78C8"}
-    col=colors.get(style,"FFFFFF")
+# ══════════════════════════════════════════════════════════════════
+#  CAPTIONS  (ffmpeg drawtext)
+# ══════════════════════════════════════════════════════════════════
+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()
+    words=caption.strip().split(); mid=max(1,len(words)//2)
+    line1=clean(" ".join(words[:mid])); line2=clean(" ".join(words[mid:])) if len(words)>1 else line1
+    col={"premium":"FFD232","energetic":"3CC8FF","fun":"FF78C8"}.get(style,"FFFFFF")
+    cta_col={"premium":"FF9900","energetic":"FF4444","fun":"AA44FF"}.get(style,"FF9900")
     out=video_path.replace(".mp4","_cap.mp4")
-
-    font_paths=[
-        "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
-        "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf",
-        "/usr/share/fonts/truetype/freefont/FreeSansBold.ttf",
-    ]
-    font=""; font_reg=""
-    for p in font_paths:
-        if os.path.exists(p): font=f":fontfile='{p}'"; font_reg=font; break
-
-    def dt(text, start, end, y, size=42, color=None, box_alpha="0.60"):
-        c = color or col
-        fd=0.4
-        return (
-            f"drawtext=text='{text}'{font}"
-            f":fontsize={size}:fontcolor=#{c}"
-            f":x=(w-text_w)/2:y={y}"
-            f":box=1:boxcolor=black@{box_alpha}:boxborderw=14"
-            f":enable='between(t,{start},{end})'"
-            f":alpha='if(lt(t,{start+fd}),(t-{start})/{fd},if(gt(t,{end-fd}),({end}-t)/{fd},1))'"
-        )
-
-    end2 = min(duration_sec-0.2, 6.5)
-
-    # 1. Main captions — inside frame, above bars
-    cap1 = dt(line1,  1.0, 3.5,  "h-190")
-    cap2 = dt(line2,  3.8, end2, "h-190")
-
-    # 2. "Shop Now" CTA — appears at 4.5s, small, bottom center
-    cta_colors={"premium":"FF9900","energetic":"FF4444","fun":"AA44FF"}
-    cta  = dt("Shop Now >", 4.5, end2, "h-130", size=32, color=cta_colors.get(style,"FF9900"), box_alpha="0.70")
-
-    # 3. Hashtag top-left — appears early
-    tag  = dt("#NewCollection", 0.5, 3.0, "60", size=28, color="FFFFFF", box_alpha="0.40")
-
-    vf = ",".join([cap1, cap2, cta, tag])
-
+    font=""
+    for p in ["/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
+              "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf"]:
+        if os.path.exists(p): font=f":fontfile='{p}'"; break
+
+    def dt(text,start,end,y,size=42,color=None,box_a="0.60"):
+        c=color or col; fd=0.4
+        return (f"drawtext=text='{text}'{font}:fontsize={size}:fontcolor=#{c}"
+                f":x=(w-text_w)/2:y={y}:box=1:boxcolor=black@{box_a}:boxborderw=14"
+                f":enable='between(t,{start},{end})'"
+                f":alpha='if(lt(t,{start+fd}),(t-{start})/{fd},if(gt(t,{end-fd}),({end}-t)/{fd},1))'")
+
+    e2=min(duration_sec-0.2,6.5)
+    vf=",".join([
+        dt(line1, 1.0, 3.5,  "h-190"),
+        dt(line2, 3.8, e2,   "h-190"),
+        dt("Shop Now >", min(4.5,e2-0.5), e2, "h-130", size=32, color=cta_col, box_a="0.70"),
+        dt("#NewCollection", 0.5, 3.0, "60", size=28, color="FFFFFF", box_a="0.40"),
+    ])
     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 — BGM + optional TTS
+#  AUDIO
 # ══════════════════════════════════════════════════════════════════
 def make_bgm(duration_sec, out_path, style="premium"):
     import wave
-    sr=44100; n=int(sr*duration_sec)
-    t=np.linspace(0,duration_sec,n,endpoint=False)
-    bpm={"premium":88,"energetic":126,"fun":104}.get(style,88)
-    beat=60./bpm
-
+    sr=44100; n=int(sr*duration_sec); t=np.linspace(0,duration_sec,n,endpoint=False)
+    bpm={"premium":88,"energetic":126,"fun":104}.get(style,88); beat=60./bpm
     kick=np.zeros(n,np.float32)
     for i in range(int(duration_sec/beat)+2):
         s=int(i*beat*sr)
         if s>=n: break
-        l=min(int(sr*.10),n-s)
-        env=np.exp(-20*np.arange(l)/sr)
+        l=min(int(sr*.10),n-s); env=np.exp(-20*np.arange(l)/sr)
         kick[s:s+l]+=env*np.sin(2*math.pi*55*np.exp(-25*np.arange(l)/sr)*np.arange(l)/sr)*0.55
-
-    bass_f={"premium":55,"energetic":80,"fun":65}.get(style,55)
-    bass=np.sin(2*math.pi*bass_f*t)*0.10*(0.5+0.5*np.sin(2*math.pi*(bpm/60/4)*t))
-
+    bf={"premium":55,"energetic":80,"fun":65}.get(style,55)
+    bass=np.sin(2*math.pi*bf*t)*0.10*(0.5+0.5*np.sin(2*math.pi*(bpm/60/4)*t))
     mf={"premium":[261,329,392],"energetic":[330,415,494],"fun":[392,494,587]}.get(style,[261,329,392])
     mel=np.zeros(n,np.float32)
     for j,f in enumerate(mf):
-        env=np.clip(0.5+0.5*np.sin(2*math.pi*1.5*t-j*2.1),0,1)
-        mel+=np.sin(2*math.pi*f*t)*env*0.045
-
+        mel+=np.sin(2*math.pi*f*t)*np.clip(0.5+0.5*np.sin(2*math.pi*1.5*t-j*2.1),0,1)*0.045
     hat=np.zeros(n,np.float32)
-    hs=beat/2
-    for i in range(int(duration_sec/hs)+2):
-        s=int(i*hs*sr)
+    for i in range(int(duration_sec/(beat/2))+2):
+        s=int(i*(beat/2)*sr)
         if s>=n: break
-        l=min(int(sr*.03),n-s)
-        hat[s:s+l]+=np.random.randn(l)*np.exp(-80*np.arange(l)/sr)*0.06
-
+        l=min(int(sr*.03),n-s); hat[s:s+l]+=np.random.randn(l)*np.exp(-80*np.arange(l)/sr)*0.06
     mix=np.clip((kick+bass+mel+hat)*0.18,-1,1)
     fade=int(sr*.5); mix[:fade]*=np.linspace(0,1,fade); mix[-fade:]*=np.linspace(1,0,fade)
-
     with wave.open(out_path,"w") as wf:
         wf.setnchannels(1); wf.setsampwidth(2); wf.setframerate(sr)
         wf.writeframes((mix*32767).astype(np.int16).tobytes())
@@ -266,197 +377,149 @@ def add_audio(video_path, caption, duration_sec, style):
     bgm=video_path.replace(".mp4","_bgm.wav")
     final=video_path.replace(".mp4","_final.mp4")
     make_bgm(duration_sec, bgm, style)
-
-    # Try TTS voiceover
     audio=bgm
     try:
         from gtts import gTTS
-        tts_mp3=video_path.replace(".mp4","_tts.mp3")
-        tts_wav=video_path.replace(".mp4","_tts.wav")
-        gTTS(text=caption[:200],lang="en",slow=False).save(tts_mp3)
+        tts=video_path.replace(".mp4","_tts.mp3"); gTTS(text=caption[:200],lang="en",slow=False).save(tts)
         mixed=video_path.replace(".mp4","_mix.wav")
-        os.system(f'ffmpeg -y -i "{bgm}" -i "{tts_mp3}" '
-                  f'-filter_complex "[0]volume=0.20[a];[1]volume=0.95[b];[a][b]amix=inputs=2:duration=first" '
+        os.system(f'ffmpeg -y -i "{bgm}" -i "{tts}" -filter_complex '
+                  f'"[0]volume=0.20[a];[1]volume=0.95[b];[a][b]amix=inputs=2:duration=first" '
                   f'-t {duration_sec} "{mixed}" -loglevel error')
         if os.path.exists(mixed): audio=mixed
-    except Exception as e: print(f"  TTS skip: {e}")
-
-    os.system(f'ffmpeg -y -i "{video_path}" -i "{audio}" '
-              f'-c:v copy -c:a aac -b:a 128k -shortest "{final}" -loglevel error')
+    except: pass
+    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
 
 
 # ══════════════════════════════════════════════════════════════════
-#  AI BRAIN — Captions, Posting Time, Target Audience
-# ══════════════════════════════════════════════════════════════════
-
-POSTING_TIMES = {
-    "Fashion":     {"slots":["7:00 AM","12:00 PM","6:00 PM","9:00 PM"],"best":"9:00 PM","days":"Tue, Thu, Fri"},
-    "Food":        {"slots":["11:00 AM","1:00 PM","7:00 PM"],"best":"12:00 PM","days":"Mon, Wed, Sat"},
-    "Tech":        {"slots":["8:00 AM","12:00 PM","5:00 PM"],"best":"8:00 AM","days":"Mon, Tue, Wed"},
-    "Beauty":      {"slots":["8:00 AM","1:00 PM","8:00 PM"],"best":"8:00 PM","days":"Wed, Fri, Sun"},
-    "Fitness":     {"slots":["6:00 AM","12:00 PM","7:00 PM"],"best":"6:00 AM","days":"Mon, Wed, Fri"},
-    "Lifestyle":   {"slots":["9:00 AM","2:00 PM","7:00 PM"],"best":"7:00 PM","days":"Thu, Fri, Sat"},
-    "Product/Other":{"slots":["10:00 AM","3:00 PM","8:00 PM"],"best":"8:00 PM","days":"Tue, Thu, Sat"},
-}
-
-AUDIENCES = {
-    "Fashion":    "👗 18-35 yo females, fashion lovers, Instagram scrollers, trend followers",
-    "Food":       "🍕 18-45 yo foodies, home cooks, restaurant goers, food bloggers",
-    "Tech":       "💻 20-40 yo tech enthusiasts, early adopters, gadget buyers, professionals",
-    "Beauty":     "💄 16-35 yo beauty lovers, skincare fans, makeup artists, self-care community",
-    "Fitness":    "💪 18-40 yo gym goers, health-conscious buyers, athletes, wellness seekers",
-    "Lifestyle":  "🌿 22-40 yo aspirational buyers, aesthetic lovers, home decor fans",
-    "Product/Other":"🛍️ 18-45 yo online shoppers, deal hunters, value-conscious buyers",
-}
-
-CAPTION_TEMPLATES = {
-    "English": {
-        "Premium":   ["{cap} ✨ Quality that speaks for itself. 🛒 Shop Now → Link in bio",
-                      "Elevate your style. {cap} 💫 DM us to order!"],
-        "Energetic": ["🔥 {cap} Hit different. Grab yours NOW 👆 Limited stock!",
-                      "⚡ Game changer alert! {cap} Don't sleep on this 🚀"],
-        "Fun":       ["Obsessed with this!! 😍 {cap} Tag someone who needs it 👇",
-                      "POV: You just found your new fav 🎉 {cap} Link in bio!"],
-    },
-    "Hindi": {
-        "Premium":   ["{cap} ✨ क्वालिटी जो बोलती है। 🛒 अभी खरीदें → Bio में link",
-                      "अपना स्टाइल बढ़ाएं। {cap} 💫 Order के लिए DM करें!"],
-        "Energetic": ["🔥 {cap} एकदम अलग है! अभी grab करो 👆 Limited stock!",
-                      "⚡ Game changer! {cap} मत सोचो, order करो 🚀"],
-        "Fun":       ["इसके साथ तो दीवाने हो जाओगे!! 😍 {cap} किसी को tag करो 👇",
-                      "POV: नया favourite मिल गया 🎉 {cap} Bio में link है!"],
-    },
-    "Hinglish": {
-        "Premium":   ["{cap} ✨ Quality toh dekho yaar! 🛒 Shop karo → Bio mein link",
-                      "Style upgrade time! {cap} 💫 DM karo order ke liye!"],
-        "Energetic": ["🔥 {cap} Bilkul alag hai bhai! Abhi lo 👆 Limited stock!",
-                      "⚡ Ek dum fire hai! {cap} Mat ruko, order karo 🚀"],
-        "Fun":       ["Yaar yeh toh kamaal hai!! 😍 {cap} Kisi ko tag karo 👇",
-                      "POV: Naya fav mil gaya 🎉 {cap} Bio mein link hai!"],
-    },
-}
-
-def detect_category(caption):
-    cap_low = caption.lower()
-    if any(w in cap_low for w in ["shoe","sneaker","dress","outfit","wear","fashion","style","cloth","jeans","kurta"]):
-        return "Fashion"
-    if any(w in cap_low for w in ["food","eat","recipe","cook","restaurant","cafe","pizza","biryani"]):
-        return "Food"
-    if any(w in cap_low for w in ["phone","laptop","tech","gadget","device","app","software","camera"]):
-        return "Tech"
-    if any(w in cap_low for w in ["skin","beauty","makeup","lipstick","cream","hair","glow","face"]):
-        return "Beauty"
-    if any(w in cap_low for w in ["gym","fit","workout","protein","yoga","health","run","sport"]):
-        return "Fitness"
-    if any(w in cap_low for w in ["home","decor","interior","lifestyle","aesthetic","plant","candle"]):
-        return "Lifestyle"
-    return "Product/Other"
-
-def get_smart_insights(caption, style, language):
-    import random, re
-    category = detect_category(caption)
-    pt = POSTING_TIMES[category]
-    audience = AUDIENCES[category]
-
-    # Generate caption in selected language
-    templates = CAPTION_TEMPLATES.get(language, CAPTION_TEMPLATES["English"])
-    style_templates = templates.get(style, templates["Premium"])
-    clean_cap = re.sub(r"[^A-Za-z0-9 !.,'-ऀ-ॿ]","",caption).strip()
-    generated_cap = random.choice(style_templates).replace("{cap}", clean_cap)
-
-    # Build insight card
-    insight = f"""📊 SMART INSIGHTS
-━━━━━━━━━━━━━━━━━━━━━━
-🎯 Category Detected: {category}
-
-👥 Target Audience:
-{audience}
-
-⏰ Best Time to Post:
-🏆 Prime Slot: {pt['best']}
-📅 Best Days: {pt['days']}
-🕐 All Good Times: {', '.join(pt['slots'])}
-
-💬 AI Caption ({language}):
-{generated_cap}
-
-#️⃣ Suggested Hashtags:
-#{category.replace('/','').replace(' ','')} #Trending #NewCollection #MustHave #ShopNow #Viral #Reels #ForYou
-━━━━━━━━━━━━━━━━━━━━━━"""
-    return insight, generated_cap
-
-
-# ══════════════════════════════════════════════════════════════════
-#  MAIN
+#  MAIN PIPELINE
 # ══════════════════════════════════════════════════════════════════
-def generate(image, caption, style, language, add_aud, add_cap, progress=gr.Progress()):
-    if image is None: return None,"⚠️ Upload an image!","Upload image first!"
-    pil=image if isinstance(image,Image.Image) else Image.fromarray(image)
-    cap=caption.strip() or "Premium Quality. Shop Now."
-    prompt=f"cinematic product ad, {cap}, smooth motion, dramatic lighting"
-    lines=[]
-    def log(msg): lines.append(msg); progress(min(.1+len(lines)*.10,.80),desc=msg)
-
-    # Get smart insights first (instant)
-    insight, ai_caption = get_smart_insights(cap, style, language)
-
-    progress(.05,desc="🚀 Generating video...")
-    video_path, model_used = get_video(pil, prompt, cb=log)
-    dur=6
-
-    # Use AI caption for video if captions enabled
-    video_caption = ai_caption if language != "English" else cap
-
-    if add_cap:
-        log("💬 Adding captions...")
-        video_path=add_captions_ffmpeg(video_path, video_caption, dur, style.lower())
-
+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."
+
+    cap = caption.strip() or ""
+    dur = int(duration)
+    lines = []
+    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.",
+            "Food":"Every bite tells a story.",
+            "Tech":"The future is here.",
+            "Beauty":"Glow different.",
+            "Fitness":"Push your limits.",
+            "Lifestyle":"Live the aesthetic.",
+            "Product/Other":"Quality that speaks for itself.",
+        }
+        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
+
+    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:
+            log(f"💬 Captions {idx+1}...")
+            video_caption = ai_cap if language != "English" else cap
+            vpath = add_captions_ffmpeg(vpath, video_caption, clip_dur, style.lower())
+
+        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...")
-        video_path=add_audio(video_path, cap, dur, style.lower())
+        final = add_audio(final, cap, dur, style.lower())
 
-    progress(1.0,desc="✅ Done!")
-    return video_path, "\n".join(lines)+f"\n\n✅ Used: {model_used}", insight
+    progress(1.0, desc="✅ Done!")
+    return final, "\n".join(lines), insight
 
 
-# ── UI ────────────────────────────────────────────────────────────
+# ══════════════════════════════════════════════════════════════════
+#  UI
+# ══════════════════════════════════════════════════════════════════
 css="""
 #title{text-align:center;font-size:2.3rem;font-weight:900}
-#sub{text-align:center;color:#888;margin-bottom:1.5rem}
-.insight{font-family:monospace;font-size:.88rem;line-height:1.7}
+#sub{text-align:center;color:#999;margin-bottom:1.2rem;font-size:1rem}
+.insight{font-family:monospace;font-size:.86rem;line-height:1.75}
 """
-with gr.Blocks(css=css,theme=gr.themes.Soft(primary_hue="violet")) as demo:
-    gr.Markdown("# 🎬 AI Reel Generator",elem_id="title")
-    gr.Markdown("Image → AI video + smart captions + posting strategy",elem_id="sub")
+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.Image(label="📸 Upload Image",type="pil",height=280)
-            cap_in  = gr.Textbox(label="✏️ Your Caption / Product Description",
-                                  value="Step into style. Own the moment.",lines=2)
+            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")
+                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 + Insights",variant="primary",size="lg")
-            gr.Markdown("**🔗 Chain:** LTX-2 ⚡ → Wan 2.2 → SVD-XT → Kling → LTX-Video → Ken Burns ✅")
+                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="🎥 Reel",height=420)
-            insight_out = gr.Textbox(label="📊 Smart Insights — Audience + Posting Time + AI Caption",
-                                      lines=18, interactive=False, elem_classes="insight")
-            log_out     = gr.Textbox(label="🔧 Log",lines=3,interactive=False)
+            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)
 
     gen_btn.click(
         fn=generate,
-        inputs=[img_in,cap_in,sty_dd,lang_dd,aud_cb,cap_cb],
-        outputs=[vid_out,log_out,insight_out],
+        inputs=[img_in, cap_in, sty_dd, lang_dd, dur_sl, aud_cb, cap_cb],
+        outputs=[vid_out, log_out, insight_out],
     )
 
-if __name__=="__main__":
+if __name__ == "__main__":
     demo.launch()