Update core/visual_engine.py

core/visual_engine.py

@@ -1,188 +1,211 @@
 # core/visual_engine.py
 from PIL import Image, ImageDraw, ImageFont, ImageOps
 # --- MONKEY PATCH FOR Image.ANTIALIAS ---
-# This is applied at module load time.
-# Attempting to make MoviePy's internal calls to Image.ANTIALIAS work with Pillow 10+
 try:
     if hasattr(Image, 'Resampling') and hasattr(Image.Resampling, 'LANCZOS'): # Pillow 9+
-        if not hasattr(Image, 'ANTIALIAS'):
-            Image.ANTIALIAS = Image.Resampling.LANCZOS
-            print("INFO: Monkey-patched PIL.Image.ANTIALIAS with Image.Resampling.LANCZOS for MoviePy compatibility.")
-    elif hasattr(Image, 'LANCZOS'): # Pillow 8 used Image.LANCZOS directly
-         if not hasattr(Image, 'ANTIALIAS'):
-            Image.ANTIALIAS = Image.LANCZOS # Pillow 8 had Image.LANCZOS
-            print("INFO: Monkey-patched PIL.Image.ANTIALIAS with Image.LANCZOS for MoviePy compatibility.")
-    else:
-        # This case means Pillow is too old, or an unexpected version.
-        # If ANTIALIAS is already present, no action needed.
-        if not hasattr(Image, 'ANTIALIAS'):
-             print("WARNING: Pillow version does not have common Resampling attributes or ANTIALIAS. Video effects might fail if MoviePy relies on ANTIALIAS.")
-except AttributeError:
-    print("WARNING: Could not perform ANTIALIAS monkey-patch due to AttributeError (Image module might be incomplete).")
-except Exception as e_mp:
-    print(f"WARNING: An unexpected error occurred during ANTIALIAS monkey-patch: {e_mp}")
+        if not hasattr(Image, 'ANTIALIAS'): Image.ANTIALIAS = Image.Resampling.LANCZOS
+    elif hasattr(Image, 'LANCZOS'): # Pillow 8
+         if not hasattr(Image, 'ANTIALIAS'): Image.ANTIALIAS = Image.LANCZOS
+    elif not hasattr(Image, 'ANTIALIAS'):
+             print("WARNING: Pillow version lacks common Resampling attributes or ANTIALIAS. Video effects might fail.")
+except Exception as e_mp: print(f"WARNING: ANTIALIAS monkey-patch error: {e_mp}")
 # --- END MONKEY PATCH ---
 
-from moviepy.editor import (ImageClip, concatenate_videoclips, TextClip, 
+from moviepy.editor import (ImageClip, VideoFileClip, concatenate_videoclips, TextClip,
                             CompositeVideoClip, AudioFileClip)
-import moviepy.video.fx.all as vfx 
+import moviepy.video.fx.all as vfx
 import numpy as np
 import os
 import openai
 import requests
 import io
 import time
-import random 
-import subprocess 
+import random
 import logging
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.INFO) # Ensure logs from this module are visible
+logger.setLevel(logging.INFO)
 
+# --- ElevenLabs Client Import ---
 ELEVENLABS_CLIENT_IMPORTED = False
-ElevenLabsAPIClient = None 
-Voice = None              
-VoiceSettings = None      
+ElevenLabsAPIClient = None
+Voice = None
+VoiceSettings = None
 try:
-    from elevenlabs.client import ElevenLabs as ImportedElevenLabsClient 
+    from elevenlabs.client import ElevenLabs as ImportedElevenLabsClient
     from elevenlabs import Voice as ImportedVoice, VoiceSettings as ImportedVoiceSettings
-    ElevenLabsAPIClient = ImportedElevenLabsClient 
+    ElevenLabsAPIClient = ImportedElevenLabsClient
     Voice = ImportedVoice
     VoiceSettings = ImportedVoiceSettings
     ELEVENLABS_CLIENT_IMPORTED = True
-    logger.info("Successfully imported ElevenLabs client components (SDK v1.x.x pattern).")
-except ImportError as e_eleven:
-    logger.warning(f"Could not import ElevenLabs client components: {e_eleven}. ElevenLabs audio will be disabled.")
-except Exception as e_gen_eleven: 
-    logger.warning(f"General error importing ElevenLabs: {e_gen_eleven}. ElevenLabs audio will be disabled.")
+    logger.info("ElevenLabs client components imported.")
+except Exception as e_eleven:
+    logger.warning(f"ElevenLabs client import failed: {e_eleven}. Audio generation 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.")
+except Exception as e_runway_sdk:
+    logger.warning(f"Error importing RunwayML SDK (placeholder): {e_runway_sdk}. RunwayML features disabled.")
 
 
 class VisualEngine:
     def __init__(self, output_dir="temp_cinegen_media", default_elevenlabs_voice_id="Rachel"):
         self.output_dir = output_dir
         os.makedirs(self.output_dir, exist_ok=True)
-        
-        self.font_filename = "arial.ttf" 
-        self.font_path_in_container = f"/usr/local/share/fonts/truetype/mycustomfonts/{self.font_filename}"
-        self.font_size_pil = 20 
-        self.video_overlay_font_size = 30 
+
+        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}"
+        ]
+        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' # More likely to be found by ImageMagick on Linux
+        self.video_overlay_font = 'Liberation-Sans-Bold' # For MoviePy TextClip
 
         try:
-            self.font = ImageFont.truetype(self.font_path_in_container, self.font_size_pil)
-            logger.info(f"Placeholder font loaded: {self.font_path_in_container}.")
+            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(f"Placeholder font '{self.font_path_in_container}' not found. Using default.")
-            self.font = ImageFont.load_default()
-            self.font_size_pil = 10
+            logger.warning("Pillow font error. 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.dalle_model = "dall-e-3"; self.image_size_dalle3 = "1792x1024"
+        self.video_frame_size = (1280, 720)
 
         self.elevenlabs_api_key = None; self.USE_ELEVENLABS = False
-        self.elevenlabs_client = None 
+        self.elevenlabs_client = None
         self.elevenlabs_voice_id = default_elevenlabs_voice_id
-        if VoiceSettings and ELEVENLABS_CLIENT_IMPORTED: 
-            self.elevenlabs_voice_settings = VoiceSettings(
-                stability=0.60, similarity_boost=0.80, 
-                style=0.15, use_speaker_boost=True
-            )
-        else:
-            self.elevenlabs_voice_settings = None
+        if VoiceSettings and ELEVENLABS_CLIENT_IMPORTED:
+            self.elevenlabs_voice_settings = VoiceSettings(stability=0.60, similarity_boost=0.80, style=0.15, use_speaker_boost=True)
+        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 >>>
+
         logger.info("VisualEngine initialized.")
 
-    def set_openai_api_key(self,k): 
+    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
-        if voice_id_from_secret: 
-            self.elevenlabs_voice_id = voice_id_from_secret
-            logger.info(f"ElevenLabs Voice ID set from config: {self.elevenlabs_voice_id}")
-        
-        if api_key and ELEVENLABS_CLIENT_IMPORTED and ElevenLabsAPIClient: 
-            try: 
-                self.elevenlabs_client = ElevenLabsAPIClient(api_key=api_key) 
-                if self.elevenlabs_client: 
-                    self.USE_ELEVENLABS=True
-                    logger.info(f"ElevenLabs Client Ready (Using Voice ID: {self.elevenlabs_voice_id}).")
-                else: 
-                    self.USE_ELEVENLABS=False; logger.warning("ElevenLabs client is None after initialization attempt.")
-            except Exception as e: 
-                logger.error(f"Error initializing ElevenLabs client: {e}. ElevenLabs Disabled.", exc_info=True); 
-                self.USE_ELEVENLABS=False; self.elevenlabs_client = None
-        else: 
-            self.USE_ELEVENLABS=False; self.elevenlabs_client = None
-            if not (ELEVENLABS_CLIENT_IMPORTED and ElevenLabsAPIClient):
-                pass # Already logged at import time if client class itself failed to import
-            else: # Client class imported, but API key was not provided
-                logger.info("ElevenLabs API Key not provided. ElevenLabs Disabled.")
-            
+        if voice_id_from_secret: self.elevenlabs_voice_id = voice_id_from_secret
+        if api_key and ELEVENLABS_CLIENT_IMPORTED and ElevenLabsAPIClient:
+            try:
+                self.elevenlabs_client = ElevenLabsAPIClient(api_key=api_key)
+                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).")
+
     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.'}")
+
+    # <<< 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
+            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 >>>
+
     def _get_text_dimensions(self,text_content,font_obj):
-        if not text_content: return 0,self.font_size_pil 
+        # ... (no changes from your previous version)
+        if not text_content: return 0,self.font_size_pil
         try:
-            if hasattr(font_obj,'getbbox'): 
+            if hasattr(font_obj,'getbbox'): # Pillow 8.0.0+
                 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'): 
+            elif hasattr(font_obj,'getsize'): # Older Pillow
                 w,h=font_obj.getsize(text_content)
                 return w, h if h > 0 else self.font_size_pil
-            else: 
+            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
-    
+
+
     def _create_placeholder_image_content(self,text_description,filename,size=None):
+        # ... (no changes from your previous version, ensure filename includes extension e.g. .png)
         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)"
         words=text_description.split();current_line=""
         for word in words:
-            test_line=current_line+word+" "; 
+            test_line=current_line+word+" ";
             if self._get_text_dimensions(test_line,self.font)[0] <= max_w: current_line=test_line
             else:
-                if current_line: lines.append(current_line.strip()); 
+                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[:max_w//int(self.font_size_pil*0.6 +1)]+"..." if text_description else "(Text too long)") # Handle single very long word
+        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
         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
 
-        y_text=padding + (size[1]-(2*padding) - max_lines_to_display*(single_line_h+2))/2.0
+        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
             if i==6 and max_lines_to_display > 7: d.text((x_text,y_text),"...",font=self.font,fill=(200,200,180));break
-        filepath=os.path.join(self.output_dir,filename); 
+        filepath=os.path.join(self.output_dir,filename);
         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)
         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"}
-        pexels_filename = output_filename_base.replace(".png", f"_pexels_{random.randint(1000,9999)}.jpg")
+        # 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:
             logger.info(f"Searching Pexels for: '{query}'"); effective_query = " ".join(query.split()[:5]); params["query"] = effective_query
             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"] 
+                photo_url = data["photos"][0]["src"]["large2x"] # Using large2x for better quality
                 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')
@@ -191,151 +214,382 @@ class VisualEngine:
         except Exception as e: logger.error(f"Pexels search/download for query '{query}': {e}", exc_info=True)
         return None
 
-    def generate_image_visual(self, image_prompt_text, scene_data, scene_identifier_filename):
-        filepath = os.path.join(self.output_dir, scene_identifier_filename)
+
+    # <<< 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
+        # --- 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.")
+        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
+        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)
+
+        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
+                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}
+            else:
+                logger.warning(f"RunwayML video clip generation failed for {base_name}. Falling back to image.")
+                # Fall through to image generation
+
+        # Image Generation (DALL-E, Pexels, Placeholder)
+        # Ensure image filename has .png
+        image_filename_with_ext = base_name + ".png"
+        filepath = os.path.join(self.output_dir, image_filename_with_ext)
+
         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) 
+                    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)
                     if revised_prompt: logger.info(f"DALL-E 3 revised_prompt: {revised_prompt[:100]}...")
                     image_response = requests.get(image_url, timeout=120); image_response.raise_for_status()
-                    img_data = Image.open(io.BytesIO(image_response.content)); 
+                    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 filepath 
-                except openai.RateLimitError as e: 
+                    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
-                    else: continue
-                except openai.APIError as e: logger.error(f"OpenAI API Error: {e}"); 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...")
+        
+        # Pexels or Placeholder if DALL-E failed or disabled
+        if self.USE_PEXELS:
             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, scene_identifier_filename)
-            if pexels_path: return pexels_path
-            logger.warning("Pexels also failed/disabled. Using placeholder.")
-            return self._create_placeholder_image_content(f"[AI/Pexels Failed] {image_prompt_text[:100]}...", scene_identifier_filename)
-        else: 
-            return self._create_placeholder_image_content(image_prompt_text, scene_identifier_filename)
+            pexels_path = self._search_pexels_image(pexels_query_text, image_filename_with_ext) # Pass filename with extension
+            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}
+
 
     def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"):
+        # ... (no changes from your previous version) ...
         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]}...")
-            
-            # Using client.text_to_speech.stream() for newer SDKs (e.g., elevenlabs >= 1.0)
+
+            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()")
-                audio_data_iterator = self.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
+                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_id=str(self.elevenlabs_voice_id), 
-                    model_id="eleven_multilingual_v2",
-                    # voice_settings=self.elevenlabs_voice_settings # Pass VoiceSettings object if supported by stream
+                    voice=voice_param,
+                    model="eleven_multilingual_v2" # or other suitable model
                 )
-            # Fallback to direct .generate() if text_to_speech.stream isn't there 
-            # AND Voice/VoiceSettings were successfully imported
-            elif hasattr(self.elevenlabs_client, 'generate') and Voice and self.elevenlabs_voice_settings:
-                 logger.info("Using elevenlabs_client.generate() with Voice object as fallback.")
-                 voice_param = Voice(voice_id=str(self.elevenlabs_voice_id), settings=self.elevenlabs_voice_settings)
-                 audio_data_iterator = 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 or Voice/VoiceSettings not imported.")
+                logger.error("No recognized audio generation method found on ElevenLabs client.")
                 return None
-            
-            with open(audio_filepath, "wb") as f:
-                for chunk in audio_data_iterator: 
-                    if chunk: f.write(chunk)
-            logger.info(f"ElevenLabs audio saved: {audio_filepath}")
-            return audio_filepath
+
+            # 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'):
+                     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
+                )
+                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 might be different.", exc_info=True)
+             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 create_video_from_images(self, image_data_list, overall_narration_path=None, output_filename="final_video.mp4", fps=24, duration_per_image=4.5):
-        if not image_data_list: logger.warning("No image data for video."); return None
-        processed_clips=[]; narration_audio_clip=None; final_video_clip_obj=None
-        logger.info(f"Preparing {len(image_data_list)} clips. Target frame: {self.video_frame_size}. Duration/img: {duration_per_image}s.")
+    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
+        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
+            scene_num = asset_info.get('scene_num', i + 1)
+            key_action = asset_info.get('key_action', '')
 
-        for i, data in enumerate(image_data_list):
-            img_path, scene_num, key_action = data.get('path'), data.get('scene_num', i+1), data.get('key_action', '')
-            if not (img_path and os.path.exists(img_path)): logger.warning(f"Img not found: {img_path}"); continue
+            if not (asset_path and os.path.exists(asset_path)):
+                logger.warning(f"Asset not found for Scene {scene_num}: {asset_path}. Skipping.")
+                continue
+            if target_scene_duration <= 0:
+                logger.warning(f"Scene {scene_num} has invalid duration ({target_scene_duration}s). Skipping.")
+                continue
+
+            current_clip = None
             try:
-                pil_img = Image.open(img_path)
-                if pil_img.mode != 'RGB': pil_img = pil_img.convert('RGB')
-                
-                img_copy = pil_img.copy()
-                img_copy.thumbnail(self.video_frame_size, Image.Resampling.LANCZOS) 
-                
-                canvas = Image.new('RGB', self.video_frame_size, (random.randint(0,5), random.randint(0,5), random.randint(0,5))) 
-                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)
+                if asset_type == 'image':
+                    pil_img = Image.open(asset_path)
+                    if pil_img.mode != 'RGB': pil_img = pil_img.convert('RGB')
+                    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)))
+                    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)
+
+                    # Ken Burns for ImageClips
+                    try:
+                        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
+
+                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.
+                    if source_video_clip.duration > target_scene_duration:
+                        current_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
+                    
+                    # 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)
+
+
+                    # 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)
+                    
+                    # 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'):
+                        source_video_clip.close()
+
+
+                else:
+                    logger.warning(f"Unknown asset type '{asset_type}' for Scene {scene_num}. Skipping.")
+                    continue
+
+                # Add text overlay
+                if current_clip and 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 = CompositeVideoClip([current_clip, txt_clip], size=self.video_frame_size, use_bgclip=True, bg_color=(0,0,0))
                 
-                img_clip_base = ImageClip(frame_np).set_duration(duration_per_image)
+                if current_clip:
+                    processed_moviepy_clips.append(current_clip)
 
-                # Ken Burns Effect (vfx.resize)
-                # The monkey patch at the top of this file should address Image.ANTIALIAS
-                img_clip = img_clip_base # Start with base
-                try:
-                    end_scale = random.uniform(1.03, 1.08); 
-                    img_clip = img_clip_base.fx(vfx.resize, lambda t: 1+(end_scale-1)*(t/duration_per_image))
-                    img_clip = img_clip.set_position('center')
-                except AttributeError as e_alias: 
-                    if 'ANTIALIAS' in str(e_alias): logger.error(f"ANTIALIAS error in vfx.resize for {img_path}. Ken Burns disabled for this clip. Error: {e_alias}")
-                    else: raise # Re-raise other AttributeErrors
-                except Exception as e_fx: logger.error(f"Error in vfx.resize for {img_path}: {e_fx}. Ken Burns disabled for this clip.")
-
-                if key_action:
-                    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(10,10,20,0.8)', method='caption', align='West',
-                                        size=(self.video_frame_size[0]*0.9, None), kerning=-1, stroke_color='black', stroke_width=1.5
-                                       ).set_duration(duration_per_image-1.0).set_start(0.5).set_position(('center',0.92),relative=True)
-                    final_scene_clip = CompositeVideoClip([img_clip, txt_clip], size=self.video_frame_size, use_bgclip=True, bg_color=(0,0,0))
-                else: final_scene_clip = img_clip
-                processed_clips.append(final_scene_clip)
-            except Exception as e: logger.error(f"Creating video clip for {img_path}: {e}", exc_info=True)
-        
-        if not processed_clips: logger.warning("No clips processed for video."); return None
-        
-        transition = 0.75 
+            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
+
+        transition_duration = 0.75
         try:
-            final_video_clip_obj = concatenate_videoclips(processed_clips, padding=-transition, method="compose")
-            if final_video_clip_obj.duration > transition*2: 
-                final_video_clip_obj = final_video_clip_obj.fx(vfx.fadein, transition).fx(vfx.fadeout, transition)
-            
+            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 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):
                 try:
                     narration_audio_clip = AudioFileClip(overall_narration_path)
-                    # Adjust video duration to match audio if audio is shorter.
-                    # If audio is longer, MoviePy's set_audio will truncate audio to video length.
-                    if narration_audio_clip.duration < final_video_clip_obj.duration:
-                        logger.info(f"Narration shorter than visuals. Trimming video to {narration_audio_clip.duration}s.")
-                        final_video_clip_obj = final_video_clip_obj.subclip(0, narration_audio_clip.duration)
+                    if final_composite_clip.duration > 0 and 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.")
                     
-                    final_video_clip_obj = final_video_clip_obj.set_audio(narration_audio_clip); logger.info("Overall narration added.")
-                except Exception as e: logger.error(f"Adding overall narration: {e}", exc_info=True)
+                    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.")
+                except Exception as e: logger.error(f"Adding narration error: {e}", exc_info=True)
             
-            output_path = os.path.join(self.output_dir, output_filename); logger.info(f"Writing final video to: {output_path}")
-            final_video_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='bar', bitrate="5000k")
-            logger.info(f"Video successfully created: {output_path}"); return output_path
-        except Exception as e: logger.error(f"Writing video file: {e}", exc_info=True); return None
+            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"
+                )
+                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
+        except Exception as e: logger.error(f"Animatic writing error: {e}", exc_info=True); return None
         finally:
-            for c_item in processed_clips: 
-                if hasattr(c_item, 'close'): c_item.close()
+            for clip in processed_moviepy_clips:
+                if hasattr(clip, 'close'): clip.close()
             if narration_audio_clip and hasattr(narration_audio_clip, 'close'): narration_audio_clip.close()
-            if final_video_clip_obj and hasattr(final_video_clip_obj, 'close'): final_video_clip_obj.close()
+            if final_composite_clip and hasattr(final_composite_clip, 'close'): final_composite_clip.close()