Update core/visual_engine.py

core/visual_engine.py

@@ -24,7 +24,7 @@ import random
 import logging
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.INFO)
+logger.setLevel(logging.INFO) # Set default logging level for this module
 
 # --- ElevenLabs Client Import ---
 ELEVENLABS_CLIENT_IMPORTED = False
@@ -38,19 +38,17 @@ try:
     Voice = ImportedVoice
     VoiceSettings = ImportedVoiceSettings
     ELEVENLABS_CLIENT_IMPORTED = True
-    logger.info("ElevenLabs client components imported.")
+    logger.info("ElevenLabs client components imported successfully.")
 except Exception as e_eleven:
-    logger.warning(f"ElevenLabs client import failed: {e_eleven}. Audio generation disabled.")
+    logger.warning(f"ElevenLabs client import failed: {e_eleven}. Audio generation will be disabled.")
 
 # --- RunwayML Client Import (Placeholder) ---
 RUNWAYML_SDK_IMPORTED = False
 RunwayMLClient = None # Placeholder for the actual RunwayML client class
 try:
-    # This is a hypothetical import. Replace with actual RunwayML SDK import if available.
     # Example: from runwayml import RunwayClient as ImportedRunwayMLClient
     # RunwayMLClient = ImportedRunwayMLClient
     # RUNWAYML_SDK_IMPORTED = True
-    # logger.info("RunwayML SDK (placeholder) imported.")
     logger.info("RunwayML SDK import is a placeholder. Actual SDK needed for Runway features.")
 except ImportError:
     logger.warning("RunwayML SDK (placeholder) not found. RunwayML video generation will be disabled.")
@@ -66,28 +64,34 @@ class VisualEngine:
         self.font_filename = "arial.ttf"
         font_paths_to_try = [
             self.font_filename,
-            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",
-            f"C:/Windows/Fonts/arial.ttf",
-            f"/usr/local/share/fonts/truetype/mycustomfonts/{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"/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
         ]
         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
+        self.video_overlay_font = 'Liberation-Sans-Bold' # For MoviePy TextClip (ImageMagick name)
 
         try:
-            self.font = ImageFont.truetype(self.font_path_pil, self.font_size_pil) if self.font_path_pil else ImageFont.load_default()
-            if self.font_path_pil: logger.info(f"Pillow font loaded: {self.font_path_pil}.")
-            else: logger.warning("Using default Pillow font."); self.font_size_pil = 10
-        except IOError:
-            logger.warning("Pillow font error. Using default."); self.font = ImageFont.load_default(); self.font_size_pil = 10
+            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
+                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.")
+            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)
+        self.video_frame_size = (1280, 720) # Standard HD 16:9
 
         self.elevenlabs_api_key = None; self.USE_ELEVENLABS = False
         self.elevenlabs_client = None
@@ -97,17 +101,14 @@ class VisualEngine:
         else: self.elevenlabs_voice_settings = None
 
         self.pexels_api_key = None; self.USE_PEXELS = False
-
-        # <<< RUNWAYML START >>>
         self.runway_api_key = None; self.USE_RUNWAYML = False
-        self.runway_client = None # Placeholder for the actual RunwayML client instance
-        # <<< RUNWAYML END >>>
+        self.runway_client = None
 
         logger.info("VisualEngine initialized.")
 
     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.'}")
+        logger.info(f"DALL-E ({self.dalle_model}) {'Ready.' if k else 'Disabled (no API key).'}")
 
     def set_elevenlabs_api_key(self,api_key, voice_id_from_secret=None):
         self.elevenlabs_api_key=api_key
@@ -118,50 +119,39 @@ 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 key or SDK).")
+        else: self.USE_ELEVENLABS=False; logger.info("ElevenLabs Disabled (no API key or SDK issue).")
 
     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.'}")
+        logger.info(f"Pexels Search {'Ready.' if k else 'Disabled (no API key).'}")
 
-    # <<< RUNWAYML START >>>
     def set_runway_api_key(self, k):
         self.runway_api_key = k
-        if k and RUNWAYML_SDK_IMPORTED and RunwayMLClient: # Assuming RunwayMLClient is the SDK's client class
+        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
-                logger.info(f"RunwayML Client (Placeholder) {'Ready.' if self.USE_RUNWAYML else 'Failed Init.'}")
-            except Exception as e:
-                logger.error(f"RunwayML client (Placeholder) init error: {e}. Disabled.", exc_info=True)
-                self.USE_RUNWAYML = False
-        elif k and not (RUNWAYML_SDK_IMPORTED and RunwayMLClient):
-            self.USE_RUNWAYML = True # Allow use with direct HTTP requests if SDK isn't used/available
-            logger.info("RunwayML API Key set. SDK (Placeholder) not imported/used. Direct API calls would be needed.")
-        else:
-            self.USE_RUNWAYML = False
-            logger.info("RunwayML Disabled (no API key or SDK issue).")
-    # <<< RUNWAYML END >>>
+                self.USE_RUNWAYML = True # Assume success for placeholder with hypothetical SDK
+                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)
+            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):
-        # ... (no changes from your previous version)
         if not text_content: return 0,self.font_size_pil
         try:
-            if hasattr(font_obj,'getbbox'): # Pillow 8.0.0+
+            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
-            elif hasattr(font_obj,'getsize'): # Older Pillow
+            elif hasattr(font_obj,'getsize'):
                 w,h=font_obj.getsize(text_content)
                 return w, h if h > 0 else self.font_size_pil
-            else: # Should not happen with standard ImageFont objects
-                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)
-        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) # Fallback
-
+            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)
+        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 changes from your previous version, ensure filename includes extension e.g. .png)
+        # (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)"
@@ -172,18 +162,14 @@ class VisualEngine:
             else:
                 if current_line: lines.append(current_line.strip());
                 current_line=word+" "
-        if current_line.strip(): lines.append(current_line.strip()) # Add last line
-        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)") # Handle single very long 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)")
         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
-        if max_lines_to_display <=0: max_lines_to_display = 1 # Ensure at least one line can be attempted
-
+        if max_lines_to_display <=0: max_lines_to_display = 1
         y_text_start = padding + (size[1]-(2*padding) - max_lines_to_display*(single_line_h+2))/2.0
         y_text = y_text_start
-
         for i in range(max_lines_to_display):
             line_content=lines[i];line_w,_=self._get_text_dimensions(line_content,self.font);x_text=(size[0]-line_w)/2.0
             d.text((x_text,y_text),line_content,font=self.font,fill=(200,200,180));y_text+=single_line_h+2
@@ -192,12 +178,10 @@ class VisualEngine:
         try:img.save(filepath);return filepath
         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 changes from your previous version, ensure output_filename_base has .png for consistency, it will be replaced)
+        # (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"}
-        # Use a more unique filename for Pexels images to avoid clashes if query is similar
         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:
@@ -205,7 +189,7 @@ class VisualEngine:
             response = requests.get("https://api.pexels.com/v1/search", headers=headers, params=params, timeout=20)
             response.raise_for_status(); data = response.json()
             if data.get("photos") and len(data["photos"]) > 0:
-                photo_url = data["photos"][0]["src"]["large2x"] # Using large2x for better quality
+                photo_url = data["photos"][0]["src"]["large2x"]
                 image_response = requests.get(photo_url, timeout=60); image_response.raise_for_status()
                 img_data = Image.open(io.BytesIO(image_response.content))
                 if img_data.mode != 'RGB': img_data = img_data.convert('RGB')
@@ -214,129 +198,61 @@ class VisualEngine:
         except Exception as e: logger.error(f"Pexels search/download for query '{query}': {e}", exc_info=True)
         return None
 
-
-    # <<< RUNWAYML START >>>
     def _generate_video_clip_with_runwayml(self, prompt_text, scene_identifier_filename_base, target_duration_seconds=4, input_image_path=None):
-        """
-        Placeholder for generating a video clip using RunwayML.
-        This needs to be implemented with the actual RunwayML SDK or API.
-        """
         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_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)")
-        logger.info(f"RunwayML Output (Placeholder): {output_video_filepath}")
-
-        # --- START ACTUAL RUNWAYML API INTERACTION (HYPOTHETICAL) ---
-        # This section is highly dependent on RunwayML's specific API/SDK.
-        # Example using a hypothetical SDK:
-        # try:
-        #     if not self.runway_client:
-        #         # self.runway_client = RunwayMLClient(api_key=self.runway_api_key) # Or however it's initialized
-        #         logger.warning("RunwayML client not initialized (Placeholder).")
-        #         # For placeholder, simulate creating a dummy video file
-        #         return self._create_placeholder_video_content(prompt_text, output_video_filename, duration=target_duration_seconds)
-
-
-        #     generation_params = {
-        #         "text_prompt": prompt_text,
-        #         "duration_seconds": target_duration_seconds,
-        #         "width": self.video_frame_size[0], # Or Runway's supported sizes
-        #         "height": self.video_frame_size[1],
-        #         # Add other params like seed, motion scale, etc.
-        #     }
-        #     if input_image_path and os.path.exists(input_image_path):
-        #         generation_params["input_image_path"] = input_image_path # For image-to-video
-        #         logger.info(f"Using input image for RunwayML: {input_image_path}")
-
-        #     task_id = self.runway_client.submit_video_generation_task(**generation_params) # Hypothetical
-        #     logger.info(f"RunwayML task submitted: {task_id}. Polling for completion...")
-
-        #     while True:
-        #         status = self.runway_client.get_task_status(task_id) # Hypothetical
-        #         if status == "completed":
-        #             video_url = self.runway_client.get_video_url(task_id) # Hypothetical
-        #             video_response = requests.get(video_url, stream=True, timeout=300)
-        #             video_response.raise_for_status()
-        #             with open(output_video_filepath, 'wb') as f:
-        #                 for chunk in video_response.iter_content(chunk_size=8192):
-        #                     f.write(chunk)
-        #             logger.info(f"RunwayML video downloaded and saved: {output_video_filepath}")
-        #             return output_video_filepath
-        #         elif status in ["failed", "error"]:
-        #             logger.error(f"RunwayML task {task_id} failed.")
-        #             return None
-        #         time.sleep(10) # Poll interval
-
-        # except Exception as e:
-        #     logger.error(f"Error during RunwayML video generation: {e}", exc_info=True)
-        #     return None
+        # --- START ACTUAL RUNWAYML API INTERACTION (HYPOTHETICAL - NEEDS IMPLEMENTATION) ---
+        # ... (Your actual RunwayML API call logic would go here) ...
         # --- END ACTUAL RUNWAYML API INTERACTION (HYPOTHETICAL) ---
-
-        # For now, as a placeholder, create a dummy MP4 file with MoviePy
-        # This allows the rest of the pipeline to be tested.
-        # **REPLACE THIS WITH ACTUAL RUNWAYML CALLS**
-        logger.warning("Using PLACEHOLDER video generation for RunwayML.")
+        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)
 
     def _create_placeholder_video_content(self, text_description, filename, duration=4, size=None):
-        """Creates a short video clip with text as a placeholder."""
         if size is None: size = self.video_frame_size
         filepath = os.path.join(self.output_dir, filename)
-        
-        # Create a simple text clip
         txt_clip = TextClip(text_description, fontsize=50, color='white', font=self.video_overlay_font,
                             bg_color='black', size=size, method='caption').set_duration(duration)
-        
         try:
             txt_clip.write_videofile(filepath, fps=24, codec='libx264', preset='ultrafast', logger=None)
             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
+        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()
-    # <<< RUNWAYML END >>>
-
 
     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):
-        """
-        Generates either an image or a video clip for a scene.
-        Returns a dictionary: {'path': asset_path, 'type': 'image'/'video', 'error': bool}
-        """
-        # Ensure scene_identifier_filename_base does not have an extension yet, or handle it
         base_name, _ = os.path.splitext(scene_identifier_filename_base)
+        asset_info = {'path': None, 'type': 'none', 'error': True, 'prompt_used': image_prompt_text, 'error_message': 'Generation not attempted'}
 
         if generate_as_video_clip and self.USE_RUNWAYML:
             logger.info(f"Attempting RunwayML video clip generation for {base_name}")
             video_path = self._generate_video_clip_with_runwayml(
-                image_prompt_text, # Use DALL-E prompt also for Runway text-to-video
-                base_name, # Pass base name, function will add .mp4
+                image_prompt_text, base_name,
                 target_duration_seconds=runway_target_duration,
                 input_image_path=input_image_for_runway
             )
             if video_path and os.path.exists(video_path):
-                return {'path': video_path, 'type': 'video', 'error': False, 'prompt_used': image_prompt_text}
+                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
 
         # Image Generation (DALL-E, Pexels, Placeholder)
-        # Ensure image filename has .png
-        image_filename_with_ext = base_name + ".png"
+        image_filename_with_ext = base_name + ".png" # Ensure .png for image
         filepath = os.path.join(self.output_dir, image_filename_with_ext)
+        asset_info['type'] = 'image' # Tentatively set type to image for this path
 
         if self.USE_AI_IMAGE_GENERATION and self.openai_api_key:
             max_retries = 2
             for attempt in range(max_retries):
                 try:
-                    # ... (DALL-E generation logic - no changes from your previous version) ...
                     logger.info(f"Attempt {attempt+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")
@@ -346,120 +262,93 @@ class VisualEngine:
                     img_data = Image.open(io.BytesIO(image_response.content));
                     if img_data.mode != 'RGB': img_data = img_data.convert('RGB')
                     img_data.save(filepath); logger.info(f"AI Image (DALL-E) saved: {filepath}");
-                    return {'path': filepath, 'type': 'image', 'error': False, 'prompt_used': image_prompt_text, 'revised_prompt': revised_prompt}
-                except openai.RateLimitError as e:
-                    logger.warning(f"OpenAI Rate Limit: {e}. Retrying after {5*(attempt+1)}s..."); time.sleep(5 * (attempt + 1))
-                    if attempt == max_retries - 1: logger.error("Max retries for RateLimitError."); break
-                except openai.APIError as e: logger.error(f"OpenAI API Error: {e}"); break
-                except requests.exceptions.RequestException as e: logger.error(f"Requests Error (DALL-E download): {e}"); break
-                except Exception as e: logger.error(f"Generic error (DALL-E gen): {e}", exc_info=True); break
-            logger.warning("DALL-E generation failed. Trying Pexels fallback...")
+                    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)
+                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...")
         
-        # Pexels or Placeholder if DALL-E failed or disabled
-        if self.USE_PEXELS:
+        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
             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) # Pass filename with extension
+            pexels_path = self._search_pexels_image(pexels_query_text, image_filename_with_ext)
             if pexels_path:
-                return {'path': pexels_path, 'type': 'image', 'error': False, 'prompt_used': f"Pexels: {pexels_query_text}"}
-            logger.warning("Pexels also failed/disabled. Using placeholder image.")
-        
-        placeholder_path = self._create_placeholder_image_content(
-            f"[AI/Pexels Failed] {image_prompt_text[:100]}...", image_filename_with_ext
-        )
-        if placeholder_path:
-            return {'path': placeholder_path, 'type': 'image', 'error': False, 'prompt_used': image_prompt_text}
-        else:
-            return {'path': None, 'type': 'none', 'error': True, 'prompt_used': image_prompt_text}
-
+                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
+
+        # 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
+            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)
 
     def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"):
-        # ... (no changes from your previous version) ...
+        # (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.")
             return None
-
         audio_filepath = os.path.join(self.output_dir, output_filename)
         try:
             logger.info(f"Generating ElevenLabs audio (Voice ID: {self.elevenlabs_voice_id}) for: {text_to_narrate[:70]}...")
-
             audio_stream_method = None
             if hasattr(self.elevenlabs_client, 'text_to_speech') and hasattr(self.elevenlabs_client.text_to_speech, 'stream'):
-                audio_stream_method = self.elevenlabs_client.text_to_speech.stream
-                logger.info("Using elevenlabs_client.text_to_speech.stream()")
-            elif hasattr(self.elevenlabs_client, 'generate_stream') : # Older SDK might have this
-                audio_stream_method = self.elevenlabs_client.generate_stream
-                logger.info("Using elevenlabs_client.generate_stream()")
-            elif hasattr(self.elevenlabs_client, 'generate'): # Fallback to non-streaming
+                audio_stream_method = self.elevenlabs_client.text_to_speech.stream; logger.info("Using elevenlabs_client.text_to_speech.stream()")
+            elif hasattr(self.elevenlabs_client, 'generate_stream') : audio_stream_method = self.elevenlabs_client.generate_stream; logger.info("Using elevenlabs_client.generate_stream()")
+            elif hasattr(self.elevenlabs_client, 'generate'):
                 logger.info("Using elevenlabs_client.generate() (non-streaming).")
-                # This one doesn't return a stream, it returns bytes directly
                 voice_param = Voice(voice_id=str(self.elevenlabs_voice_id), settings=self.elevenlabs_voice_settings) if Voice and self.elevenlabs_voice_settings else str(self.elevenlabs_voice_id)
-                audio_bytes = self.elevenlabs_client.generate(
-                    text=text_to_narrate,
-                    voice=voice_param,
-                    model="eleven_multilingual_v2" # or other suitable model
-                )
-                with open(audio_filepath, "wb") as f:
-                    f.write(audio_bytes)
-                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
+                audio_bytes = self.elevenlabs_client.generate(text=text_to_narrate, voice=voice_param, model="eleven_multilingual_v2")
+                with open(audio_filepath, "wb") as f: f.write(audio_bytes)
+                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 we have a streaming method
             if audio_stream_method:
                 voice_param_for_stream = {"voice_id": str(self.elevenlabs_voice_id)}
-                # For Pydantic v1 style for elevenlabs sdk <1.0
-                # if self.elevenlabs_voice_settings and hasattr(self.elevenlabs_voice_settings, 'dict'):
-                #    voice_param_for_stream["voice_settings"] = self.elevenlabs_voice_settings.dict()
-                # For Pydantic v2 style for elevenlabs skd >=1.0
-                if self.elevenlabs_voice_settings and hasattr(self.elevenlabs_voice_settings, 'model_dump'):
+                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 : # If not a pydantic model, pass as is if supported
-                     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
-                )
+                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
+                
+                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:
                     for chunk in audio_data_iterator:
                         if chunk: f.write(chunk)
-                logger.info(f"ElevenLabs audio (streamed) saved: {audio_filepath}")
-                return audio_filepath
-
-        except AttributeError as ae:
-             logger.error(f"AttributeError with ElevenLabs client: {ae}. SDK method/params might be different.", exc_info=True)
-        except Exception as e:
-            logger.error(f"Error generating ElevenLabs audio: {e}", exc_info=True)
+                logger.info(f"ElevenLabs audio (streamed) saved: {audio_filepath}"); return audio_filepath
+        except AttributeError as ae: logger.error(f"AttributeError with ElevenLabs client: {ae}. SDK method/params might be different.", exc_info=True)
+        except Exception as e: logger.error(f"Error generating ElevenLabs audio: {e}", exc_info=True)
         return None
 
     def assemble_animatic_from_assets(self, asset_data_list, overall_narration_path=None, output_filename="final_video.mp4", fps=24):
-        """
-        Assembles the final video from a list of assets (images or video clips).
-        Each item in asset_data_list should be a dict like:
-        {'path': 'path/to/asset', 'type': 'image'|'video', 'duration': desired_scene_duration_in_animatic,
-         'scene_num': num, 'key_action': 'text'}
-        """
         if not asset_data_list:
             logger.warning("No asset data provided for animatic assembly.")
             return None
 
         processed_moviepy_clips = []
         narration_audio_clip = None
-        final_composite_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.")
 
         for i, asset_info in enumerate(asset_data_list):
             asset_path = asset_info.get('path')
             asset_type = asset_info.get('type')
-            # This 'duration' is the desired display duration of THIS scene in the final animatic
-            target_scene_duration = asset_info.get('duration', 4.5) # Default if not specified
+            target_scene_duration = asset_info.get('duration', 4.5)
             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")
+
             if not (asset_path and os.path.exists(asset_path)):
                 logger.warning(f"Asset not found for Scene {scene_num}: {asset_path}. Skipping.")
                 continue
@@ -467,65 +356,82 @@ class VisualEngine:
                 logger.warning(f"Scene {scene_num} has invalid duration ({target_scene_duration}s). Skipping.")
                 continue
 
-            current_clip = None
+            current_clip_for_scene = None
             try:
                 if asset_type == 'image':
+                    logger.debug(f"S{scene_num}: Loading image asset from {asset_path}")
                     pil_img = Image.open(asset_path)
-                    if pil_img.mode != 'RGB': pil_img = pil_img.convert('RGB')
+                    logger.debug(f"S{scene_num}: Image loaded. 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
+
                     img_copy = pil_img.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)
-                    canvas = Image.new('RGB', self.video_frame_size, (random.randint(0,10), random.randint(0,10), random.randint(0,10)))
+                    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.paste(img_copy, (xo, yo))
-                    frame_np = np.array(canvas)
-                    current_clip_base = ImageClip(frame_np).set_duration(target_scene_duration)
+                    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.")
 
-                    # Ken Burns for ImageClips
-                    try:
+                    # 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}")
+
+                    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
+                    
+                    current_clip_base = ImageClip(frame_np, transparent=False, ismask=False).set_duration(target_scene_duration)
+                    logger.debug(f"S{scene_num}: Base ImageClip created.")
+
+                    current_clip_for_scene = current_clip_base
+                    try: # Ken Burns
                         end_scale = random.uniform(1.03, 1.08)
-                        current_clip = current_clip_base.fx(vfx.resize, lambda t: 1 + (end_scale - 1) * (t / target_scene_duration)).set_position('center')
-                    except Exception as e_fx:
-                        logger.error(f"Ken Burns error for image {asset_path}: {e_fx}. Using static image.")
-                        current_clip = current_clip_base
+                        current_clip_for_scene = current_clip_base.fx(vfx.resize, lambda t: 1 + (end_scale - 1) * (t / target_scene_duration)).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
 
                 elif asset_type == 'video':
-                    source_video_clip = VideoFileClip(asset_path, target_resolution=(self.video_frame_size[1], self.video_frame_size[0]))
-                    # Fit video into target_scene_duration:
-                    # If source is shorter, it will play once. If longer, it will be cut.
-                    # For more complex looping/speed adjustments, more logic is needed.
+                    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:
-                        current_clip = source_video_clip.subclip(0, target_scene_duration)
+                        temp_clip = source_video_clip.subclip(0, target_scene_duration)
                     elif source_video_clip.duration < target_scene_duration:
-                         # Simple loop if significantly shorter, or just play once if close
-                        if target_scene_duration / source_video_clip.duration > 1.5 and source_video_clip.duration > 0.1 : # Loop if target is >150% of source
-                             current_clip = source_video_clip.loop(duration=target_scene_duration)
-                        else: # Play once, duration will be its own, MoviePy handles concatenation padding
-                             current_clip = source_video_clip.set_duration(source_video_clip.duration) # Explicitly set
-                             logger.info(f"Runway clip for S{scene_num} ({source_video_clip.duration:.2f}s) shorter than target ({target_scene_duration:.2f}s), will play once.")
-                    else: # Durations match
-                        current_clip = source_video_clip
+                        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.")
                     
-                    # Ensure the clip has the target duration for consistent concatenation
-                    if current_clip.duration != target_scene_duration:
-                         current_clip = current_clip.set_duration(target_scene_duration)
-
+                    # Crucially, ensure the clip used in concatenation has the target_scene_duration
+                    current_clip_for_scene = temp_clip.set_duration(target_scene_duration)
 
-                    # Resize if necessary (MoviePy does this on CompositeVideoClip too, but explicit can be good)
-                    if current_clip.size != list(self.video_frame_size):
-                        current_clip = current_clip.resize(self.video_frame_size)
+                    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)
                     
-                    # Close the original source_video_clip if it's different from current_clip (e.g., after subclip)
-                    if current_clip != source_video_clip and hasattr(source_video_clip, 'close'):
+                    # 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
 
-                else:
-                    logger.warning(f"Unknown asset type '{asset_type}' for Scene {scene_num}. Skipping.")
-                    continue
-
-                # Add text overlay
-                if current_clip and key_action:
+                if current_clip_for_scene and key_action: # Add text overlay
+                    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:
@@ -535,61 +441,48 @@ class VisualEngine:
                                             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 = CompositeVideoClip([current_clip, txt_clip], size=self.video_frame_size, use_bgclip=True, bg_color=(0,0,0))
+                        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.")
                 
-                if current_clip:
-                    processed_moviepy_clips.append(current_clip)
+                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()
 
-            except Exception as e:
-                logger.error(f"Error processing asset for Scene {scene_num} ({asset_path}): {e}", exc_info=True)
-                if current_clip and hasattr(current_clip, 'close'): current_clip.close() # Ensure closure on error
-
-        if not processed_moviepy_clips:
-            logger.warning("No MoviePy clips successfully processed. Aborting animatic assembly.")
-            return None
 
+        if not processed_moviepy_clips: logger.warning("No MoviePy clips 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: # Should have been caught above, but defensive
-                logger.error("No clips available for final concatenation.")
-                return None
+            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 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 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 overall_narration_path and os.path.exists(overall_narration_path):
+            if overall_narration_path and os.path.exists(overall_narration_path) and final_composite_clip.duration > 0:
                 try:
                     narration_audio_clip = AudioFileClip(overall_narration_path)
-                    if final_composite_clip.duration > 0 and narration_audio_clip.duration < final_composite_clip.duration:
+                    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)
-                    elif final_composite_clip.duration <= 0: logger.warning("Video has no duration. Audio not added.")
-                    
-                    if narration_audio_clip and final_composite_clip.duration > 0: # Check again
-                        final_composite_clip = final_composite_clip.set_audio(narration_audio_clip)
-                        logger.info("Overall narration added.")
+                    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.")
             
             if final_composite_clip and final_composite_clip.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"
-                )
+                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 or has no duration. Not writing file."); return None
+            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
         finally:
-            for clip in processed_moviepy_clips:
-                if hasattr(clip, 'close'): clip.close()
+            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()