Update app.py

app.py

@@ -1,13 +1,12 @@
 """
-AI Reels Maker — HuggingFace Space  v3.1
+AI Reels Maker — HuggingFace Space  v3.3
 =========================================
-Professional viral reel generator with:
-• 45-60 second reels (8+ sentences, 15-20 words each)
-• 2-3 stitched background videos from Pexels
-• Optional outro video (plays after main reel)
-• Editable script: generate, edit, then render
-• Factual, CTA‑enforced prompts (no invented data)
-• Robust Edge TTS with retry & fallback
+- Clean captions (no background, no shadow)
+- Reliable Pexels background videos
+- Voice selection works (Edge TTS & Kokoro)
+- Optional outro video
+- Factual, CTA‑enforced prompts
+- Groq & Pexels API secrets support
 """
 
 import sys
@@ -40,7 +39,7 @@ FONT_BOLD_PATH  = "/tmp/Montserrat-Bold.ttf"
 FONT_LIGHT_PATH = "/tmp/Montserrat-Regular.ttf"
 
 # ─────────────────────────────────────────────────────────────────────────────
-# VOICE CONFIGS
+# VOICE CONFIGS (plain ASCII keys)
 # ─────────────────────────────────────────────────────────────────────────────
 EDGE_VOICES = {
     "Aria (US Female)"      : "en-US-AriaNeural",
@@ -234,7 +233,7 @@ def generate_script(
     return lines[:num_points]
 
 # ─────────────────────────────────────────────────────────────────────────────
-# AUDIO — Edge TTS (improved with retry and fallback)
+# AUDIO — Edge TTS (with voice selection debug)
 # ─────────────────────────────────────────────────────────────────────────────
 async def _edge_tts_save(text: str, voice: str, path: str) -> None:
     import edge_tts
@@ -246,16 +245,18 @@ def generate_audio_edge(text: str, voice_key: str) -> str:
     if len(text.strip()) < 5:
         raise ValueError(f"Text too short: '{text}'")
 
+    # Get voice ID from key; if not found, use default
     voice = EDGE_VOICES.get(voice_key)
     if voice is None:
-        print(f"[TTS] Unknown voice key '{voice_key}', using default '{DEFAULT_VOICE_KEY}'")
+        print(f"[TTS] WARNING: Voice key '{voice_key}' not found. Using default '{DEFAULT_VOICE_KEY}'")
         voice = EDGE_VOICES[DEFAULT_VOICE_KEY]
+    else:
+        print(f"[TTS] Using voice: {voice} (key: {voice_key})")
 
-    print(f"[TTS] Using voice ID: {voice} (key: {voice_key})")
     print(f"[TTS] Text snippet: {text[:200]}... (total chars: {len(text)})")
 
     out = tempfile.mktemp(suffix=".mp3")
-    # Try selected voice first, then fallback to default voice
+    # Try selected voice first, then fallback to default voice if it fails
     for attempt, try_voice in enumerate([voice, EDGE_VOICES[DEFAULT_VOICE_KEY]]):
         if attempt > 0:
             print(f"[TTS] Retry with fallback voice: {try_voice}")
@@ -299,7 +300,8 @@ def generate_audio_kokoro(text: str, voice_key: str) -> str:
     if _kokoro_pipeline is None:
         from kokoro import KPipeline
         _kokoro_pipeline = KPipeline(lang_code="a")
-    voice  = KOKORO_VOICES.get(voice_key, "af_heart")
+    voice = KOKORO_VOICES.get(voice_key, "af_heart")
+    print(f"[Kokoro] Using voice: {voice}")
     chunks = [a for _, _, a in _kokoro_pipeline(text, voice=voice, speed=1.05)]
     out    = tempfile.mktemp(suffix=".wav")
     sf.write(out, np.concatenate(chunks), 24000)
@@ -347,7 +349,8 @@ def _fetch_one_video(query: str, api_key: str, used_ids: set) -> str | None:
             timeout=20,
         )
         resp.raise_for_status()
-        for vid in resp.json().get("videos", []):
+        data = resp.json()
+        for vid in data.get("videos", []):
             vid_id = vid.get("id")
             if vid_id in used_ids:
                 continue
@@ -371,14 +374,23 @@ def _fetch_one_video(query: str, api_key: str, used_ids: set) -> str | None:
     return None
 
 def fetch_bg_videos(sentences: list[str], api_key: str, count: int = 3) -> list[str]:
-    queries   = _extract_queries(sentences, count=count)
-    used_ids  = set()
-    paths     = []
+    if not api_key:
+        print("[pexels] No API key provided – skipping background videos.")
+        return []
+    queries = _extract_queries(sentences, count=count)
+    print(f"[pexels] Searching for videos with queries: {queries}")
+    used_ids = set()
+    paths = []
     for q in queries:
         p = _fetch_one_video(q, api_key, used_ids)
         if p:
             paths.append(p)
-    print(f"[pexels] Fetched {len(paths)}/{count} videos", flush=True)
+            if len(paths) >= count:
+                break
+    if not paths:
+        print("[pexels] WARNING: No videos found. Using gradient fallback.")
+    else:
+        print(f"[pexels] Successfully fetched {len(paths)} videos: {paths}")
     return paths
 
 # ─────────────────────────────────────────────────────────────────────────────
@@ -413,7 +425,7 @@ def paste_logo(base: Image.Image, logo: Image.Image | None, position: str = "top
     return out
 
 # ─────────────────────────────────────────────────────────────────────────────
-# TEXT FRAME RENDERER
+# TEXT FRAME RENDERER (NO background, NO shadow)
 # ─────────────────────────────────────────────────────────────────────────────
 def render_text_frame(
     text: str, width: int, height: int,
@@ -421,15 +433,17 @@ def render_text_frame(
     logo_pos: str = "top-right",
     accent_color: tuple = (124, 58, 237),
 ) -> np.ndarray:
-    img  = Image.new("RGBA", (width, height), (0, 0, 0, 0))
+    """Render text with NO background and NO shadow – just clean text on video."""
+    img = Image.new("RGBA", (width, height), (0, 0, 0, 0))  # fully transparent
     draw = ImageDraw.Draw(img)
     font = get_font(60, bold=True)
 
+    # Word wrap to fit 90% of width (no background, so more space)
     words, lines, cur = text.split(), [], []
     for word in words:
         test = " ".join(cur + [word])
         bbox = draw.textbbox((0, 0), test, font=font)
-        if bbox[2] - bbox[0] > width * 0.84 and cur:
+        if bbox[2] - bbox[0] > width * 0.90 and cur:
             lines.append(" ".join(cur))
             cur = [word]
         else:
@@ -437,26 +451,14 @@ def render_text_frame(
     if cur:
         lines.append(" ".join(cur))
 
-    line_h       = 76
-    total_text_h = len(lines) * line_h
-    pad_v, pad_h = 44, 24
-    box_y1 = max(pad_h, (height - total_text_h) // 2 - pad_v)
-    box_y2 = min(height - pad_h, box_y1 + total_text_h + pad_v * 2)
-
-    overlay = Image.new("RGBA", (width, height), (0, 0, 0, 0))
-    od      = ImageDraw.Draw(overlay)
-    od.rounded_rectangle([20, box_y1, width - 20, box_y2], radius=22, fill=(6, 6, 20, 190))
-    od.rectangle([20, box_y1, 29, box_y2], fill=(*accent_color, 245))
-    img  = Image.alpha_composite(img, overlay)
-    draw = ImageDraw.Draw(img)
+    line_h = 72
+    y = (height - (len(lines) * line_h)) // 2
 
-    y = box_y1 + pad_v
     for line in lines:
-        bbox   = draw.textbbox((0, 0), line, font=font)
+        bbox = draw.textbbox((0, 0), line, font=font)
         text_w = bbox[2] - bbox[0]
-        x      = (width - text_w) // 2
-        for dx, dy in [(-2, -2), (2, -2), (-2, 2), (2, 2), (0, 3), (3, 0), (-3, 0), (0, -3)]:
-            draw.text((x + dx, y + dy), line, font=font, fill=(0, 0, 0, 170))
+        x = (width - text_w) // 2
+        # Draw text directly, no shadow, no background
         draw.text((x, y), line, font=font, fill=(255, 255, 255, 255))
         y += line_h
 
@@ -529,7 +531,8 @@ def create_reel(
                     loops = int(np.ceil(seg_dur / bg.duration)) + 1
                     bg = mpe.concatenate_videoclips([bg] * loops)
                 bg = bg.subclip(0, seg_dur)
-                dark = mpe.ColorClip((W, H), color=[0, 0, 0]).set_opacity(0.42).set_duration(seg_dur)
+                # Darken only if you want, but now text has no background so we keep it lighter
+                dark = mpe.ColorClip((W, H), color=[0, 0, 0]).set_opacity(0.35).set_duration(seg_dur)
                 seg = mpe.CompositeVideoClip([bg, dark]).set_start(start)
                 bg_segs.append(seg)
                 continue
@@ -540,7 +543,7 @@ def create_reel(
 
     bg_layer = mpe.CompositeVideoClip(bg_segs, size=(W, H)).set_duration(total_dur)
 
-    # Text overlays
+    # Text overlays (clean, no background)
     text_clips = []
     for i, sentence in enumerate(sentences):
         arr = render_text_frame(sentence, W, H, logo, logo_pos, accent_color)
@@ -559,7 +562,7 @@ def create_reel(
     return out
 
 # ─────────────────────────────────────────────────────────────────────────────
-# NEWS ANCHOR MODE (simplified, kept from previous working version)
+# NEWS ANCHOR MODE (simplified)
 # ─────────────────────────────────────────────────────────────────────────────
 def _make_studio_bg(W: int, H: int, dark: bool, accent_color: tuple) -> np.ndarray:
     frame = np.zeros((H, W, 3), dtype=np.uint8)
@@ -876,7 +879,7 @@ with gr.Blocks(title="🎬 AI Reels Maker", css=CSS, theme=gr.themes.Soft()) as
                     logo_file = gr.File(label="📤 Upload Logo (PNG/JPG)", file_types=["image"], type="filepath")
                     logo_pos  = gr.Dropdown(choices=["top-right","top-left","bottom-right","bottom-left"], value="top-right", label="Logo Position")
                     accent_hex = gr.ColorPicker(value="#d4af37", label="Accent Color")
-                    show_caption_bg = gr.Checkbox(label="Show caption background", value=True, info="Toggle semi-transparent card behind text")
+                    show_caption_bg = gr.Checkbox(label="Show caption background", value=False, info="Unchecked = no background/shadow (clean text)", visible=False)  # hidden because we removed background
                     outro_video = gr.File(label="🎬 Optional Outro Video (plays after the reel)", file_types=["video"], type="filepath")
 
                 num_points = gr.Slider(6, 10, value=8, step=1, label="📊 Number of points (8 = ~55s reel)")