Update core/visual_engine.py

core/visual_engine.py

@@ -61,37 +61,40 @@ class VisualEngine:
         self.output_dir = output_dir
         os.makedirs(self.output_dir, exist_ok=True)
 
-        self.font_filename = "arial.ttf"
+        self.font_filename = "arial.ttf" # Or a more reliably found font like "DejaVuSans-Bold.ttf"
         font_paths_to_try = [
             self.font_filename,
-            f"/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", # Common on Linux
-            f"/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf", # Common on Linux
+            f"/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
+            f"/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf",
             f"/System/Library/Fonts/Supplemental/Arial.ttf", # macOS
             f"C:/Windows/Fonts/arial.ttf", # Windows
-            f"/usr/local/share/fonts/truetype/mycustomfonts/{self.font_filename}" # Custom container path
+            f"/usr/local/share/fonts/truetype/mycustomfonts/{self.font_filename}"
         ]
         self.font_path_pil = next((p for p in font_paths_to_try if os.path.exists(p)), None)
         self.font_size_pil = 20
         self.video_overlay_font_size = 30
         self.video_overlay_font_color = 'white'
-        self.video_overlay_font = 'Liberation-Sans-Bold' # For MoviePy TextClip (ImageMagick name)
+        # For MoviePy TextClip, use font names ImageMagick knows. Check with `convert -list font`.
+        # 'Liberation-Sans-Bold' is a good default if available.
+        self.video_overlay_font = 'DejaVuSans-Bold' if 'dejavu' in (self.font_path_pil or '').lower() else 'Liberation-Sans-Bold'
+
 
         try:
             if self.font_path_pil:
                 self.font = ImageFont.truetype(self.font_path_pil, self.font_size_pil)
                 logger.info(f"Pillow font loaded: {self.font_path_pil}.")
-            else: # Fallback to default if no path found
+            else:
                 self.font = ImageFont.load_default()
-                logger.warning("Custom Pillow font not found from paths. Using default. Text rendering might be basic.")
-                self.font_size_pil = 10 # Default font is smaller
-        except IOError as e_font: # Catch specific IOError for font loading
-            logger.error(f"Pillow font loading IOError for '{self.font_path_pil if self.font_path_pil else 'default'}': {e_font}. Using default.")
+                logger.warning("Custom Pillow font not found. Using default. Text rendering for placeholders might be basic.")
+                self.font_size_pil = 10 # Default Pillow font is small
+        except IOError as e_font:
+            logger.error(f"Pillow font loading IOError for '{self.font_path_pil or 'default'}': {e_font}. Using default.")
             self.font = ImageFont.load_default()
             self.font_size_pil = 10
 
         self.openai_api_key = None; self.USE_AI_IMAGE_GENERATION = False
         self.dalle_model = "dall-e-3"; self.image_size_dalle3 = "1792x1024"
-        self.video_frame_size = (1280, 720) # Standard HD 16:9
+        self.video_frame_size = (1280, 720)
 
         self.elevenlabs_api_key = None; self.USE_ELEVENLABS = False
         self.elevenlabs_client = None
@@ -108,7 +111,7 @@ class VisualEngine:
 
     def set_openai_api_key(self,k):
         self.openai_api_key=k; self.USE_AI_IMAGE_GENERATION=bool(k)
-        logger.info(f"DALL-E ({self.dalle_model}) {'Ready.' if k else 'Disabled (no API key).'}")
+        logger.info(f"DALL-E ({self.dalle_model}) {'Ready.' if k else 'Disabled.'}")
 
     def set_elevenlabs_api_key(self,api_key, voice_id_from_secret=None):
         self.elevenlabs_api_key=api_key
@@ -119,39 +122,38 @@ class VisualEngine:
                 self.USE_ELEVENLABS=bool(self.elevenlabs_client)
                 logger.info(f"ElevenLabs Client {'Ready' if self.USE_ELEVENLABS else 'Failed Init'} (Voice ID: {self.elevenlabs_voice_id}).")
             except Exception as e: logger.error(f"ElevenLabs client init error: {e}. Disabled.", exc_info=True); self.USE_ELEVENLABS=False
-        else: self.USE_ELEVENLABS=False; logger.info("ElevenLabs Disabled (no API key or SDK issue).")
+        else: self.USE_ELEVENLABS=False; logger.info("ElevenLabs Disabled (no key or SDK).")
 
     def set_pexels_api_key(self,k):
         self.pexels_api_key=k; self.USE_PEXELS=bool(k)
-        logger.info(f"Pexels Search {'Ready.' if k else 'Disabled (no API key).'}")
+        logger.info(f"Pexels Search {'Ready.' if k else 'Disabled.'}")
 
     def set_runway_api_key(self, k):
         self.runway_api_key = k
         if k and RUNWAYML_SDK_IMPORTED and RunwayMLClient:
             try:
                 # self.runway_client = RunwayMLClient(api_key=k) # Actual initialization
-                self.USE_RUNWAYML = True # Assume success for placeholder with hypothetical SDK
+                self.USE_RUNWAYML = True
                 logger.info(f"RunwayML Client (Placeholder with SDK) {'Ready.' if self.USE_RUNWAYML else 'Failed Init.'}")
             except Exception as e: logger.error(f"RunwayML client (Placeholder with SDK) init error: {e}. Disabled.", exc_info=True); self.USE_RUNWAYML = False
-        elif k: # API key provided, but SDK might not be used/imported (e.g., direct HTTP)
+        elif k:
             self.USE_RUNWAYML = True
             logger.info("RunwayML API Key set. Using direct API calls or placeholder (SDK not fully integrated/imported).")
         else: self.USE_RUNWAYML = False; logger.info("RunwayML Disabled (no API key).")
 
     def _get_text_dimensions(self,text_content,font_obj):
-        if not text_content: return 0,self.font_size_pil
+        if not text_content: return 0, (self.font.size if hasattr(self.font, 'size') else self.font_size_pil)
         try:
             if hasattr(font_obj,'getbbox'):
                 bbox=font_obj.getbbox(text_content);w=bbox[2]-bbox[0];h=bbox[3]-bbox[1]
-                return w, h if h > 0 else self.font_size_pil
+                return w, h if h > 0 else font_obj.size
             elif hasattr(font_obj,'getsize'):
                 w,h=font_obj.getsize(text_content)
-                return w, h if h > 0 else self.font_size_pil
-            else: return int(len(text_content)*self.font_size_pil*0.6),int(self.font_size_pil*1.2 if self.font_size_pil*1.2>0 else self.font_size_pil)
+                return w, h if h > 0 else font_obj.size
+            else: return int(len(text_content)*font_obj.size*0.6), int(font_obj.size*1.2)
         except Exception as e: logger.warning(f"Error in _get_text_dimensions for '{text_content[:20]}...': {e}"); return int(len(text_content)*self.font_size_pil*0.6),int(self.font_size_pil*1.2)
 
     def _create_placeholder_image_content(self,text_description,filename,size=None):
-        # (No significant changes from your previous correct version)
         if size is None: size = self.video_frame_size
         img=Image.new('RGB',size,color=(20,20,40));d=ImageDraw.Draw(img);padding=25;max_w=size[0]-(2*padding);lines=[];
         if not text_description: text_description="(Placeholder: No prompt text)"
@@ -163,7 +165,7 @@ class VisualEngine:
                 if current_line: lines.append(current_line.strip());
                 current_line=word+" "
         if current_line.strip(): lines.append(current_line.strip())
-        if not lines and text_description: lines.append(text_description[:int(max_w//(self.font_size_pil*0.6 +1))]+"..." if text_description else "(Text too long)")
+        if not lines and text_description: lines.append(text_description[:int(max_w//(self._get_text_dimensions("A",self.font)[0] or 10))]+"..." if text_description else "(Text too long)")
         elif not lines: lines.append("(Placeholder Text Error)")
         _,single_line_h=self._get_text_dimensions("Ay",self.font); single_line_h = single_line_h if single_line_h > 0 else self.font_size_pil + 2
         max_lines_to_display=min(len(lines),(size[1]-(2*padding))//(single_line_h+2)) if single_line_h > 0 else 1
@@ -179,9 +181,8 @@ class VisualEngine:
         except Exception as e:logger.error(f"Saving placeholder image {filepath}: {e}", exc_info=True);return None
 
     def _search_pexels_image(self, query, output_filename_base):
-        # (No significant changes from your previous correct version)
         if not self.USE_PEXELS or not self.pexels_api_key: return None
-        headers = {"Authorization": self.pexels_api_key}; params = {"query": query, "per_page": 1, "orientation": "landscape", "size": "large"}
+        headers = {"Authorization": self.pexels_api_key}; params = {"query": query, "per_page": 1, "orientation": "landscape", "size": "large2x"} # Request higher quality
         pexels_filename = output_filename_base.replace(".png", f"_pexels_{random.randint(1000,9999)}.jpg").replace(".mp4", f"_pexels_{random.randint(1000,9999)}.jpg")
         filepath = os.path.join(self.output_dir, pexels_filename)
         try:
@@ -202,11 +203,20 @@ class VisualEngine:
         if not self.USE_RUNWAYML or not self.runway_api_key:
             logger.warning("RunwayML not enabled or API key missing. Cannot generate video clip.")
             return None
-        output_video_filename = scene_identifier_filename_base.replace(".png", ".mp4") # Ensure .mp4 extension
+        output_video_filename = scene_identifier_filename_base.replace(".png", "_runway.mp4") # More specific extension
         output_video_filepath = os.path.join(self.output_dir, output_video_filename)
         logger.info(f"Attempting RunwayML video generation for: {prompt_text[:100]}... (Target duration: {target_duration_seconds}s)")
         # --- START ACTUAL RUNWAYML API INTERACTION (HYPOTHETICAL - NEEDS IMPLEMENTATION) ---
-        # ... (Your actual RunwayML API call logic would go here) ...
+        # Example:
+        # if self.runway_client:
+        #     try:
+        #         # result = self.runway_client.generate(text=prompt_text, duration=target_duration_seconds, seed_image=input_image_path)
+        #         # result.save(output_video_filepath)
+        #         # return output_video_filepath
+        #     except Exception as e_runway:
+        #         logger.error(f"Actual RunwayML generation error: {e_runway}", exc_info=True)
+        #         return None
+        # else: logger.warning("RunwayML client not initialized (placeholder).")
         # --- END ACTUAL RUNWAYML API INTERACTION (HYPOTHETICAL) ---
         logger.warning("Using PLACEHOLDER video generation for RunwayML as actual API calls are not implemented.")
         return self._create_placeholder_video_content(f"[RunwayML Placeholder] {prompt_text}", output_video_filename, duration=target_duration_seconds)
@@ -214,15 +224,16 @@ class VisualEngine:
     def _create_placeholder_video_content(self, text_description, filename, duration=4, size=None):
         if size is None: size = self.video_frame_size
         filepath = os.path.join(self.output_dir, filename)
-        txt_clip = TextClip(text_description, fontsize=50, color='white', font=self.video_overlay_font,
-                            bg_color='black', size=size, method='caption').set_duration(duration)
+        txt_clip = None # Initialize
         try:
-            txt_clip.write_videofile(filepath, fps=24, codec='libx264', preset='ultrafast', logger=None)
+            txt_clip = TextClip(text_description, fontsize=50, color='white', font=self.video_overlay_font,
+                                bg_color='black', size=size, method='caption').set_duration(duration)
+            txt_clip.write_videofile(filepath, fps=24, codec='libx264', preset='ultrafast', logger=None, threads=2)
             logger.info(f"Placeholder video saved: {filepath}")
             return filepath
         except Exception as e: logger.error(f"Failed to create placeholder video {filepath}: {e}", exc_info=True); return None
         finally:
-            if hasattr(txt_clip, 'close'): txt_clip.close()
+            if txt_clip and hasattr(txt_clip, 'close'): txt_clip.close()
 
     def generate_scene_asset(self, image_prompt_text, scene_data, scene_identifier_filename_base,
                              generate_as_video_clip=False, runway_target_duration=4, input_image_for_runway=None):
@@ -238,22 +249,18 @@ class VisualEngine:
             )
             if video_path and os.path.exists(video_path):
                 asset_info = {'path': video_path, 'type': 'video', 'error': False, 'prompt_used': image_prompt_text}
-                return asset_info # Successfully generated video
-            else:
-                logger.warning(f"RunwayML video clip generation failed for {base_name}. Falling back to image.")
-                asset_info['error_message'] = "RunwayML video generation failed."
-                # Fall through to image generation
+                return asset_info
+            else: logger.warning(f"RunwayML video clip generation failed for {base_name}. Falling back to image."); asset_info['error_message'] = "RunwayML video generation failed."
 
-        # Image Generation (DALL-E, Pexels, Placeholder)
-        image_filename_with_ext = base_name + ".png" # Ensure .png for image
+        image_filename_with_ext = base_name + ".png"
         filepath = os.path.join(self.output_dir, image_filename_with_ext)
-        asset_info['type'] = 'image' # Tentatively set type to image for this path
+        asset_info['type'] = 'image'
 
         if self.USE_AI_IMAGE_GENERATION and self.openai_api_key:
-            max_retries = 2
-            for attempt in range(max_retries):
+            max_retries = 2; attempt_num = 0
+            for attempt_num in range(max_retries):
                 try:
-                    logger.info(f"Attempt {attempt+1}: DALL-E ({self.dalle_model}) for: {image_prompt_text[:100]}...")
+                    logger.info(f"Attempt {attempt_num+1}: DALL-E ({self.dalle_model}) for: {image_prompt_text[:100]}...")
                     client = openai.OpenAI(api_key=self.openai_api_key, timeout=90.0)
                     response = client.images.generate(model=self.dalle_model, prompt=image_prompt_text, n=1, size=self.image_size_dalle3, quality="hd", response_format="url", style="vivid")
                     image_url = response.data[0].url; revised_prompt = getattr(response.data[0], 'revised_prompt', None)
@@ -263,39 +270,37 @@ class VisualEngine:
                     if img_data.mode != 'RGB': img_data = img_data.convert('RGB')
                     img_data.save(filepath); logger.info(f"AI Image (DALL-E) saved: {filepath}");
                     asset_info = {'path': filepath, 'type': 'image', 'error': False, 'prompt_used': image_prompt_text, 'revised_prompt': revised_prompt}
-                    return asset_info
-                except openai.RateLimitError as e_rate: logger.warning(f"OpenAI Rate Limit: {e_rate}. Retrying..."); time.sleep(5 * (attempt + 1)); asset_info['error_message'] = str(e_rate)
+                    return asset_info # Success
+                except openai.RateLimitError as e_rate: logger.warning(f"OpenAI Rate Limit on attempt {attempt_num+1}: {e_rate}. Retrying..."); time.sleep(5 * (attempt_num + 1)); asset_info['error_message'] = str(e_rate)
                 except openai.APIError as e_api: logger.error(f"OpenAI API Error: {e_api}"); asset_info['error_message'] = str(e_api); break
                 except requests.exceptions.RequestException as e_req: logger.error(f"Requests Error (DALL-E download): {e_req}"); asset_info['error_message'] = str(e_req); break
                 except Exception as e_gen: logger.error(f"Generic error (DALL-E gen): {e_gen}", exc_info=True); asset_info['error_message'] = str(e_gen); break
-                if attempt == max_retries - 1: logger.error("Max retries for DALL-E RateLimitError."); break
-            if asset_info['error']: logger.warning("DALL-E generation failed. Trying Pexels fallback...")
+            if asset_info['error']: logger.warning(f"DALL-E generation failed after {attempt_num+1} attempts. Trying Pexels fallback...")
         
-        if self.USE_PEXELS and (asset_info['error'] or not (self.USE_AI_IMAGE_GENERATION and self.openai_api_key)): # Try Pexels if DALL-E failed or disabled
+        if self.USE_PEXELS and (asset_info['error'] or not (self.USE_AI_IMAGE_GENERATION and self.openai_api_key)):
             pexels_query_text = scene_data.get('pexels_search_query_감독', f"{scene_data.get('emotional_beat','')} {scene_data.get('setting_description','')}")
             pexels_path = self._search_pexels_image(pexels_query_text, image_filename_with_ext)
             if pexels_path:
                 asset_info = {'path': pexels_path, 'type': 'image', 'error': False, 'prompt_used': f"Pexels: {pexels_query_text}"}
                 return asset_info
-            asset_info['error_message'] = (asset_info.get('error_message', "") + " Pexels search also failed or disabled.").strip()
-            if not asset_info['error']: logger.warning("Pexels search failed or disabled.") # If DALL-E wasn't even tried
+            current_error_msg = asset_info.get('error_message', "")
+            asset_info['error_message'] = (current_error_msg + " Pexels search also failed or disabled.").strip()
+            if not asset_info['error']: logger.warning("Pexels search failed or was disabled (DALL-E not attempted).")
 
-        # Fallback to placeholder if all else fails
-        if asset_info['error']: # Only create placeholder if previous steps failed
-            logger.warning("All generation methods failed. Using placeholder image.")
-            placeholder_prompt_text = asset_info.get('prompt_used', image_prompt_text) # Use the prompt that was attempted
+        if asset_info['error']:
+            logger.warning("All primary generation methods failed. Using placeholder image.")
+            placeholder_prompt_text = asset_info.get('prompt_used', image_prompt_text)
             placeholder_path = self._create_placeholder_image_content(f"[Fallback Placeholder] {placeholder_prompt_text[:100]}...", image_filename_with_ext)
             if placeholder_path:
                 asset_info = {'path': placeholder_path, 'type': 'image', 'error': False, 'prompt_used': placeholder_prompt_text}
-                return asset_info
-            else: # Final failure
-                asset_info['error_message'] = (asset_info.get('error_message', "") + " Placeholder creation also failed.").strip()
-        return asset_info # Return whatever state asset_info is in (could be error=True)
+            else:
+                current_error_msg = asset_info.get('error_message', "")
+                asset_info['error_message'] = (current_error_msg + " Placeholder creation also failed.").strip()
+        return asset_info
 
     def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"):
-        # (No significant changes from your previous correct version, ensure error handling is robust)
         if not self.USE_ELEVENLABS or not self.elevenlabs_client or not text_to_narrate:
-            logger.info("ElevenLabs conditions not met (API key, client init, or text). Skipping audio.")
+            logger.info("ElevenLabs conditions not met. Skipping audio generation.")
             return None
         audio_filepath = os.path.join(self.output_dir, output_filename)
         try:
@@ -312,13 +317,12 @@ class VisualEngine:
                 logger.info(f"ElevenLabs audio (non-streamed) saved: {audio_filepath}"); return audio_filepath
             else: logger.error("No recognized audio generation method found on ElevenLabs client."); return None
 
-            if audio_stream_method:
+            if audio_stream_method: # Streaming logic
                 voice_param_for_stream = {"voice_id": str(self.elevenlabs_voice_id)}
-                if self.elevenlabs_voice_settings and hasattr(self.elevenlabs_voice_settings, 'model_dump'): # Pydantic v2 for elevenlabs sdk >=1.0
-                     voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings.model_dump()
-                elif self.elevenlabs_voice_settings and hasattr(self.elevenlabs_voice_settings, 'dict'): # Pydantic v1 for elevenlabs sdk <1.0
-                     voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings.dict()
-                elif self.elevenlabs_voice_settings : voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings
+                if self.elevenlabs_voice_settings:
+                    if hasattr(self.elevenlabs_voice_settings, 'model_dump'): voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings.model_dump() # Pydantic v2
+                    elif hasattr(self.elevenlabs_voice_settings, 'dict'): voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings.dict() # Pydantic v1
+                    else: voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings
                 
                 audio_data_iterator = audio_stream_method(text=text_to_narrate, model_id="eleven_multilingual_v2", **voice_param_for_stream)
                 with open(audio_filepath, "wb") as f:
@@ -336,153 +340,191 @@ class VisualEngine:
 
         processed_moviepy_clips = []
         narration_audio_clip = None
-        final_composite_clip = None # Renamed to avoid conflict in finally block
-        total_video_duration_from_assets = sum(item.get('duration', 4.5) for item in asset_data_list)
-        logger.info(f"Assembling animatic from {len(asset_data_list)} assets. Target frame: {self.video_frame_size}. Approx total duration: {total_video_duration_from_assets:.2f}s.")
+        final_composite_clip_obj = None
+        
+        logger.info(f"Assembling animatic from {len(asset_data_list)} assets. Target frame: {self.video_frame_size}.")
 
         for i, asset_info in enumerate(asset_data_list):
             asset_path = asset_info.get('path')
             asset_type = asset_info.get('type')
-            target_scene_duration = asset_info.get('duration', 4.5)
+            target_scene_duration = asset_info.get('duration', 4.5) # Duration for this scene in the animatic
             scene_num = asset_info.get('scene_num', i + 1)
             key_action = asset_info.get('key_action', '')
 
-            logger.info(f"Processing Scene {scene_num}: Path='{asset_path}', Type='{asset_type}', Target Duration='{target_scene_duration}'s")
+            logger.info(f"Processing S{scene_num}: Path='{asset_path}', Type='{asset_type}', TargetDur='{target_scene_duration}'s")
 
             if not (asset_path and os.path.exists(asset_path)):
-                logger.warning(f"Asset not found for Scene {scene_num}: {asset_path}. Skipping.")
-                continue
+                logger.warning(f"S{scene_num}: Asset not found at '{asset_path}'. Skipping."); continue
             if target_scene_duration <= 0:
-                logger.warning(f"Scene {scene_num} has invalid duration ({target_scene_duration}s). Skipping.")
-                continue
+                logger.warning(f"S{scene_num}: Invalid duration ({target_scene_duration}s). Skipping."); continue
 
-            current_clip_for_scene = None
+            current_scene_clip = None # The final MoviePy clip for this scene
             try:
                 if asset_type == 'image':
-                    logger.debug(f"S{scene_num}: Loading image asset from {asset_path}")
                     pil_img = Image.open(asset_path)
-                    logger.debug(f"S{scene_num}: Image loaded. Mode: {pil_img.mode}, Size: {pil_img.size}")
+                    logger.debug(f"S{scene_num}: Loaded image. Mode: {pil_img.mode}, Size: {pil_img.size}")
 
-                    # Ensure image is RGBA for consistent pasting, then convert to RGB for MoviePy
-                    if pil_img.mode != 'RGBA':
-                        pil_img = pil_img.convert('RGBA') # Convert to RGBA to handle transparency uniformly
+                    # 1. Ensure image is RGBA for consistent alpha handling during processing
+                    img_rgba_source = pil_img.convert('RGBA') if pil_img.mode != 'RGBA' else pil_img.copy()
 
-                    img_copy = pil_img.copy()
+                    # 2. Thumbnail the RGBA image
+                    img_thumbnail = img_rgba_source.copy() # Work on a copy
                     resample_filter = Image.Resampling.LANCZOS if hasattr(Image.Resampling, 'LANCZOS') else (Image.ANTIALIAS if hasattr(Image, 'ANTIALIAS') else Image.BILINEAR)
-                    img_copy.thumbnail(self.video_frame_size, resample_filter)
-                    logger.debug(f"S{scene_num}: Image thumbnailed to: {img_copy.size}")
-
-                    # Create an RGBA canvas, paste the (potentially RGBA) image onto it
-                    canvas_rgba = Image.new('RGBA', self.video_frame_size, (0, 0, 0, 0)) # Fully transparent
-                    xo, yo = (self.video_frame_size[0] - img_copy.width) // 2, (self.video_frame_size[1] - img_copy.height) // 2
-                    canvas_rgba.paste(img_copy, (xo, yo), img_copy) # Paste using image's own alpha
-                    logger.debug(f"S{scene_num}: Image pasted onto RGBA canvas.")
-
-                    # Now create a final RGB canvas and paste the RGBA canvas onto it, effectively blending alpha
-                    final_rgb_canvas = Image.new("RGB", self.video_frame_size, (random.randint(0,5), random.randint(0,5), random.randint(0,5))) # Dark background
-                    final_rgb_canvas.paste(canvas_rgba, mask=canvas_rgba.split()[3]) # Use alpha channel of canvas_rgba as mask
-                    
-                    debug_canvas_path = os.path.join(self.output_dir, f"debug_final_rgb_canvas_scene_{scene_num}.png")
-                    try: final_rgb_canvas.save(debug_canvas_path); logger.info(f"DEBUG: Saved final RGB canvas for scene {scene_num} to {debug_canvas_path}")
-                    except Exception as e_save_canvas: logger.error(f"DEBUG: Failed to save final RGB canvas for scene {scene_num}: {e_save_canvas}")
+                    img_thumbnail.thumbnail(self.video_frame_size, resample_filter)
+                    logger.debug(f"S{scene_num}: Thumbnailed to: {img_thumbnail.size}")
 
-                    frame_np = np.array(final_rgb_canvas)
-                    logger.debug(f"S{scene_num}: Final RGB canvas to NumPy. Shape: {frame_np.shape}, Dtype: {frame_np.dtype}")
-                    if frame_np.size == 0: logger.error(f"S{scene_num}: NumPy array for ImageClip is empty! Skipping."); continue
+                    # 3. Create a target-sized RGBA canvas (fully transparent)
+                    canvas_rgba = Image.new('RGBA', self.video_frame_size, (0, 0, 0, 0))
                     
-                    current_clip_base = ImageClip(frame_np, transparent=False, ismask=False).set_duration(target_scene_duration)
-                    logger.debug(f"S{scene_num}: Base ImageClip created.")
+                    # 4. Paste the thumbnailed image (with its alpha) onto the center of the RGBA canvas
+                    xo = (self.video_frame_size[0] - img_thumbnail.width) // 2
+                    yo = (self.video_frame_size[1] - img_thumbnail.height) // 2
+                    canvas_rgba.paste(img_thumbnail, (xo, yo), img_thumbnail) # Use img_thumbnail's alpha as mask
+                    logger.debug(f"S{scene_num}: Image pasted onto transparent RGBA canvas.")
+
+                    # 5. Create a final RGB image by pasting the RGBA canvas onto an opaque background
+                    #    This flattens transparency and ensures an RGB image for MoviePy.
+                    final_rgb_image_for_moviepy = Image.new("RGB", self.video_frame_size, (0, 0, 0)) # Opaque black background
+                    final_rgb_image_for_moviepy.paste(canvas_rgba, mask=canvas_rgba.split()[3]) # Paste using alpha from canvas_rgba
+
+                    # --- CRITICAL DEBUG STEP: Save the image that will be fed to MoviePy ---
+                    debug_canvas_path = os.path.join(self.output_dir, f"debug_final_rgb_FOR_MOVIEPY_scene_{scene_num}.png")
+                    try:
+                        final_rgb_image_for_moviepy.save(debug_canvas_path)
+                        logger.info(f"DEBUG: Saved final RGB image for MoviePy (S{scene_num}) to {debug_canvas_path}")
+                    except Exception as e_save_canvas:
+                        logger.error(f"DEBUG: Failed to save final_rgb_image_for_moviepy (S{scene_num}): {e_save_canvas}")
+                    
+                    frame_np = np.array(final_rgb_image_for_moviepy) # Should be (H, W, 3) dtype uint8
+                    logger.debug(f"S{scene_num}: Converted to NumPy. Shape: {frame_np.shape}, Dtype: {frame_np.dtype}, Size: {frame_np.size}")
+
+                    if frame_np.size == 0: logger.error(f"S{scene_num}: NumPy array is EMPTY. Skipping."); continue
+                    if frame_np.ndim != 3 or frame_np.shape[2] != 3: logger.error(f"S{scene_num}: NumPy array has unexpected shape {frame_np.shape}. Skipping."); continue
+                    if frame_np.dtype != np.uint8: frame_np = frame_np.astype(np.uint8); logger.warning(f"S{scene_num}: Converted NumPy array dtype to uint8.")
 
-                    current_clip_for_scene = current_clip_base
+                    current_clip_base = ImageClip(frame_np, transparent=False).set_duration(target_scene_duration)
+                    logger.debug(f"S{scene_num}: Base ImageClip created from NumPy array.")
+                    
+                    current_scene_clip_with_fx = current_clip_base # Start with base
                     try: # Ken Burns
                         end_scale = random.uniform(1.03, 1.08)
-                        current_clip_for_scene = current_clip_base.fx(vfx.resize, lambda t: 1 + (end_scale - 1) * (t / target_scene_duration)).set_position('center')
+                        current_scene_clip_with_fx = current_clip_base.fx(vfx.resize, lambda t: 1 + (end_scale - 1) * (t / target_scene_duration) if target_scene_duration > 0 else 1).set_position('center')
                         logger.debug(f"S{scene_num}: Ken Burns effect applied.")
-                    except Exception as e_fx: logger.error(f"S{scene_num}: Ken Burns error: {e_fx}. Using static.", exc_info=False); current_clip_for_scene = current_clip_base
+                    except Exception as e_fx: logger.error(f"S{scene_num}: Ken Burns error: {e_fx}. Using static.", exc_info=False)
+                    
+                    current_scene_clip = current_scene_clip_with_fx
 
                 elif asset_type == 'video':
                     logger.debug(f"S{scene_num}: Loading video asset from {asset_path}")
-                    # Ensure target_resolution is (height, width) for VideoFileClip resizing parameter
-                    source_video_clip = VideoFileClip(asset_path, target_resolution=(self.video_frame_size[1], self.video_frame_size[0]) if self.video_frame_size else None)
-                    
-                    temp_clip = source_video_clip # Work with a temporary variable
-                    if source_video_clip.duration > target_scene_duration:
-                        temp_clip = source_video_clip.subclip(0, target_scene_duration)
-                    elif source_video_clip.duration < target_scene_duration:
-                        if target_scene_duration / source_video_clip.duration > 1.5 and source_video_clip.duration > 0.1:
-                             temp_clip = source_video_clip.loop(duration=target_scene_duration)
-                        else: # Play once, MoviePy will pad if needed during concatenation if durations differ
-                             temp_clip = source_video_clip.set_duration(source_video_clip.duration) # Keep its own duration
-                             logger.info(f"Video clip for S{scene_num} ({source_video_clip.duration:.2f}s) is shorter than target animatic duration ({target_scene_duration:.2f}s). It will play once at its native length.")
-                    
-                    # Crucially, ensure the clip used in concatenation has the target_scene_duration
-                    current_clip_for_scene = temp_clip.set_duration(target_scene_duration)
+                    source_video_clip = None # Initialize
+                    try:
+                        source_video_clip = VideoFileClip(asset_path, target_resolution=(self.video_frame_size[1], self.video_frame_size[0]) if self.video_frame_size else None)
+                        
+                        temp_clip_for_video_asset = source_video_clip
+                        if source_video_clip.duration != target_scene_duration:
+                            if source_video_clip.duration > target_scene_duration:
+                                temp_clip_for_video_asset = source_video_clip.subclip(0, target_scene_duration)
+                            else: # Source is shorter
+                                if target_scene_duration / source_video_clip.duration > 1.5 and source_video_clip.duration > 0.1:
+                                    temp_clip_for_video_asset = source_video_clip.loop(duration=target_scene_duration)
+                                else: # Let it play its native length, will be set to target_scene_duration for concat
+                                    temp_clip_for_video_asset = source_video_clip.set_duration(source_video_clip.duration) 
+                                    logger.info(f"S{scene_num}: Video clip ({source_video_clip.duration:.2f}s) shorter than scene target ({target_scene_duration:.2f}s).")
+                        
+                        current_scene_clip = temp_clip_for_video_asset.set_duration(target_scene_duration)
+
+                        if current_scene_clip.size != list(self.video_frame_size):
+                            logger.debug(f"S{scene_num}: Resizing video clip from {current_scene_clip.size} to {self.video_frame_size}")
+                            current_scene_clip = current_scene_clip.resize(self.video_frame_size)
+                        
+                        logger.debug(f"S{scene_num}: Video asset processed. Final duration for scene: {current_scene_clip.duration:.2f}s")
+                    except Exception as e_vid_load:
+                        logger.error(f"S{scene_num}: Error loading/processing video file '{asset_path}': {e_vid_load}", exc_info=True)
+                        if source_video_clip and hasattr(source_video_clip, 'close'): source_video_clip.close()
+                        continue # Skip this asset
+                    finally: # Close original source if it was opened and different from the final clip
+                         if source_video_clip and source_video_clip is not current_scene_clip and hasattr(source_video_clip, 'close'):
+                             source_video_clip.close()
 
-                    if current_clip_for_scene.size != list(self.video_frame_size):
-                        logger.debug(f"S{scene_num}: Resizing video clip from {current_clip_for_scene.size} to {self.video_frame_size}")
-                        current_clip_for_scene = current_clip_for_scene.resize(self.video_frame_size)
-                    
-                    # Only close source_video_clip if it's different from what we are keeping (e.g., after subclip)
-                    # And if it's not the same object as current_clip_for_scene
-                    if source_video_clip is not current_clip_for_scene and hasattr(source_video_clip, 'close'):
-                        source_video_clip.close()
-                    logger.debug(f"S{scene_num}: Video asset processed. Final duration for scene: {current_clip_for_scene.duration:.2f}s")
 
                 else: logger.warning(f"S{scene_num}: Unknown asset type '{asset_type}'. Skipping."); continue
-
-                if current_clip_for_scene and key_action: # Add text overlay
+                
+                # Add text overlay (common to both image and video assets)
+                if current_scene_clip and key_action:
                     logger.debug(f"S{scene_num}: Adding text overlay: '{key_action}'")
                     text_overlay_duration = min(target_scene_duration - 0.5, target_scene_duration * 0.8) if target_scene_duration > 0.5 else target_scene_duration
                     text_overlay_start = (target_scene_duration - text_overlay_duration) / 2.0
                     if text_overlay_duration > 0:
-                        txt_clip = TextClip(f"Scene {scene_num}\n{key_action}",
-                                            fontsize=self.video_overlay_font_size, color=self.video_overlay_font_color,
-                                            font=self.video_overlay_font, bg_color='rgba(10,10,20,0.7)',
-                                            method='caption', align='West', size=(self.video_frame_size[0] * 0.9, None),
-                                            kerning=-1, stroke_color='black', stroke_width=1.5
-                                           ).set_duration(text_overlay_duration).set_start(text_overlay_start).set_position(('center', 0.92), relative=True)
-                        current_clip_for_scene = CompositeVideoClip([current_clip_for_scene, txt_clip], size=self.video_frame_size, use_bgclip=True)
-                        logger.debug(f"S{scene_num}: Text overlay composited.")
+                        try:
+                            txt_clip = TextClip(f"Scene {scene_num}\n{key_action}",
+                                                fontsize=self.video_overlay_font_size, color=self.video_overlay_font_color,
+                                                font=self.video_overlay_font, bg_color='rgba(10,10,20,0.7)',
+                                                method='caption', align='West', size=(self.video_frame_size[0] * 0.9, None),
+                                                kerning=-1, stroke_color='black', stroke_width=1.5
+                                               ).set_duration(text_overlay_duration).set_start(text_overlay_start).set_position(('center', 0.92), relative=True)
+                            current_scene_clip = CompositeVideoClip([current_scene_clip, txt_clip], size=self.video_frame_size, use_bgclip=True)
+                            logger.debug(f"S{scene_num}: Text overlay composited.")
+                        except Exception as e_txt: logger.error(f"S{scene_num}: Error creating TextClip or CompositeVideoClip for text: {e_txt}. Using clip without text.", exc_info=True)
                 
-                if current_clip_for_scene: processed_moviepy_clips.append(current_clip_for_scene); logger.info(f"S{scene_num}: Asset successfully processed and added to final list.")
-            except Exception as e: logger.error(f"Error processing asset for Scene {scene_num} ({asset_path}): {e}", exc_info=True)
-            finally: # Ensure individual clips are closed if they were opened and an error occurred mid-processing
-                if current_clip_for_scene and asset_type == 'video' and hasattr(current_clip_for_scene, 'reader') and current_clip_for_scene.reader:
-                    if hasattr(current_clip_for_scene, 'close'): current_clip_for_scene.close()
-
-
-        if not processed_moviepy_clips: logger.warning("No MoviePy clips processed. Aborting animatic assembly."); return None
+                if current_scene_clip:
+                    processed_moviepy_clips.append(current_scene_clip)
+                    logger.info(f"S{scene_num}: Asset successfully processed. Clip duration: {current_scene_clip.duration:.2f}s, Added to final list.")
+
+            except Exception as e_asset_proc:
+                logger.error(f"MAJOR Error processing asset for Scene {scene_num} ({asset_path}): {e_asset_proc}", exc_info=True)
+                # Ensure clip is closed if it was partially created
+                if current_scene_clip and hasattr(current_scene_clip, 'reader') and current_scene_clip.reader:
+                    if hasattr(current_scene_clip, 'close'): current_scene_clip.close()
+                elif current_scene_clip and hasattr(current_scene_clip, 'close'):
+                    current_scene_clip.close()
+
+        if not processed_moviepy_clips: logger.warning("No MoviePy clips were successfully processed. Aborting animatic assembly."); return None
         
         transition_duration = 0.75
         try:
-            if len(processed_moviepy_clips) > 1: final_composite_clip = concatenate_videoclips(processed_moviepy_clips, padding=-transition_duration, method="compose")
-            elif processed_moviepy_clips: final_composite_clip = processed_moviepy_clips[0]
-            else: logger.error("No clips for final concatenation."); return None
+            if not processed_moviepy_clips: logger.error("No clips to concatenate after processing loop."); return None
+            logger.info(f"Concatenating {len(processed_moviepy_clips)} processed clips.")
+            if len(processed_moviepy_clips) > 1:
+                final_composite_clip_obj = concatenate_videoclips(processed_moviepy_clips, padding = -transition_duration if transition_duration > 0 else 0, method="compose")
+            elif processed_moviepy_clips: final_composite_clip_obj = processed_moviepy_clips[0]
+            
+            if not final_composite_clip_obj: logger.error("Concatenation resulted in a None clip."); return None
+            logger.info(f"Concatenated clip duration: {final_composite_clip_obj.duration:.2f}s")
 
-            if final_composite_clip.duration > transition_duration * 2: final_composite_clip = final_composite_clip.fx(vfx.fadein, transition_duration).fx(vfx.fadeout, transition_duration)
-            elif final_composite_clip.duration > 0: final_composite_clip = final_composite_clip.fx(vfx.fadein, min(transition_duration, final_composite_clip.duration/2.0))
+            if transition_duration > 0:
+                if final_composite_clip_obj.duration > transition_duration * 2:
+                    final_composite_clip_obj = final_composite_clip_obj.fx(vfx.fadein, transition_duration).fx(vfx.fadeout, transition_duration)
+                elif final_composite_clip_obj.duration > 0:
+                    final_composite_clip_obj = final_composite_clip_obj.fx(vfx.fadein, min(transition_duration, final_composite_clip_obj.duration/2.0))
+                logger.debug("Applied fade in/out effects.")
 
-            if overall_narration_path and os.path.exists(overall_narration_path) and final_composite_clip.duration > 0:
+            if overall_narration_path and os.path.exists(overall_narration_path) and final_composite_clip_obj.duration > 0:
                 try:
                     narration_audio_clip = AudioFileClip(overall_narration_path)
-                    if narration_audio_clip.duration < final_composite_clip.duration:
-                        logger.info(f"Narration ({narration_audio_clip.duration:.2f}s) shorter than visuals ({final_composite_clip.duration:.2f}s). Trimming video.")
-                        final_composite_clip = final_composite_clip.subclip(0, narration_audio_clip.duration)
-                    final_composite_clip = final_composite_clip.set_audio(narration_audio_clip); logger.info("Overall narration added.")
-                except Exception as e: logger.error(f"Adding narration error: {e}", exc_info=True)
-            elif final_composite_clip.duration <= 0 : logger.warning("Video has no duration. Audio not added.")
+                    logger.info(f"Adding narration. Video dur: {final_composite_clip_obj.duration:.2f}s, Audio dur: {narration_audio_clip.duration:.2f}s")
+                    final_composite_clip_obj = final_composite_clip_obj.set_audio(narration_audio_clip) # Audio will be cut/padded to video duration
+                    logger.info("Overall narration added to video.")
+                except Exception as e_audio: logger.error(f"Error adding overall narration: {e_audio}", exc_info=True)
+            elif final_composite_clip_obj.duration <= 0 : logger.warning("Video has no duration. Audio not added.")
             
-            if final_composite_clip and final_composite_clip.duration > 0:
+            if final_composite_clip_obj and final_composite_clip_obj.duration > 0:
                 output_path = os.path.join(self.output_dir, output_filename)
-                logger.info(f"Writing final animatic: {output_path} (Duration: {final_composite_clip.duration:.2f}s)")
-                final_composite_clip.write_videofile(output_path, fps=fps, codec='libx264', preset='medium', audio_codec='aac',
-                                            temp_audiofile=os.path.join(self.output_dir, f'temp-audio-{os.urandom(4).hex()}.m4a'),
-                                            remove_temp=True, threads=os.cpu_count() or 2, logger='bar', bitrate="5000k")
-                logger.info(f"Animatic created: {output_path}"); return output_path
-            else: logger.error("Final animatic clip invalid. Not writing file."); return None
-        except Exception as e: logger.error(f"Animatic writing error: {e}", exc_info=True); return None
+                logger.info(f"Attempting to write final animatic: {output_path} (Duration: {final_composite_clip_obj.duration:.2f}s)")
+                moviepy_logger_setting = 'bar' # Default to progress bar
+
+                final_composite_clip_obj.write_videofile(
+                    output_path, fps=fps, codec='libx264', preset='medium', audio_codec='aac',
+                    temp_audiofile=os.path.join(self.output_dir, f'temp-audio-{os.urandom(4).hex()}.m4a'),
+                    remove_temp=True, threads=os.cpu_count() or 2, logger=moviepy_logger_setting, bitrate="5000k"
+                )
+                logger.info(f"Animatic video successfully created: {output_path}")
+                return output_path
+            else: logger.error("Final animatic clip is invalid or has zero duration. Cannot write file."); return None
+        except Exception as e_write: logger.error(f"Error during video file writing or final composition: {e_write}", exc_info=True); return None
         finally:
-            for clip_obj in processed_moviepy_clips:
-                if hasattr(clip_obj, 'close'): clip_obj.close()
-            if narration_audio_clip and hasattr(narration_audio_clip, 'close'): narration_audio_clip.close()
-            if final_composite_clip and hasattr(final_composite_clip, 'close'): final_composite_clip.close()
+            logger.debug("Closing all MoviePy clips in `assemble_animatic_from_assets` finally block.")
+            clips_to_close = processed_moviepy_clips + ([narration_audio_clip] if narration_audio_clip else []) + ([final_composite_clip_obj] if final_composite_clip_obj else [])
+            for clip_obj in clips_to_close:
+                if clip_obj and hasattr(clip_obj, 'close'):
+                    try: clip_obj.close()
+                    except Exception as e_close: logger.warning(f"Ignoring error while closing a clip: {e_close}")