# core/visual_engine.py from PIL import Image, ImageDraw, ImageFont, ImageOps import base64 import mimetypes import numpy as np import os import openai import requests import io import time import random import logging # --- MoviePy Imports --- from moviepy.editor import ( ImageClip, VideoFileClip, concatenate_videoclips, TextClip, CompositeVideoClip, AudioFileClip, ) import moviepy.video.fx.all as vfx # --- MONKEY PATCH for Pillow/MoviePy compatibility --- try: if hasattr(Image, "Resampling") and hasattr(Image.Resampling, "LANCZOS"): # Pillow 9+ 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"): # Fallback if no common resampling attributes found print( "WARNING: Pillow version lacks common Resampling attributes or ANTIALIAS. MoviePy effects might fail or look different." ) except Exception as e_monkey_patch: print( f"WARNING: An unexpected error occurred during Pillow ANTIALIAS monkey-patch: {e_monkey_patch}" ) logger = logging.getLogger(__name__) # Consider setting level in main app if not already configured: # logger.setLevel(logging.DEBUG) # For very verbose output during debugging # --- External Service Client Imports --- ELEVENLABS_CLIENT_IMPORTED = False ElevenLabsAPIClient = None Voice = None VoiceSettings = None try: from elevenlabs.client import ElevenLabs as ImportedElevenLabsClient from elevenlabs import Voice as ImportedVoice, VoiceSettings as ImportedVoiceSettings ElevenLabsAPIClient = ImportedElevenLabsClient Voice = ImportedVoice VoiceSettings = ImportedVoiceSettings ELEVENLABS_CLIENT_IMPORTED = True logger.info("ElevenLabs client components imported successfully.") except ImportError: logger.warning( "ElevenLabs SDK not found (pip install elevenlabs). Audio generation will be disabled." ) except Exception as e_eleven_import: logger.warning( f"Error importing ElevenLabs client components: {e_eleven_import}. Audio generation disabled." ) RUNWAYML_SDK_IMPORTED = False RunwayMLAPIClient = None # Using a more specific name for the client class try: from runwayml import RunwayML as ImportedRunwayMLClient # Actual SDK import RunwayMLAPIClient = ImportedRunwayMLClient RUNWAYML_SDK_IMPORTED = True logger.info("RunwayML SDK imported successfully.") except ImportError: logger.warning( "RunwayML SDK not found (pip install runwayml). RunwayML video generation will be disabled." ) except Exception as e_runway_sdk_import: logger.warning( f"Error importing RunwayML SDK: {e_runway_sdk_import}. RunwayML features disabled." ) class VisualEngine: DEFAULT_FONT_SIZE_PIL = 10 # For default Pillow font PREFERRED_FONT_SIZE_PIL = 20 # For custom font VIDEO_OVERLAY_FONT_SIZE = 30 VIDEO_OVERLAY_FONT_COLOR = "white" # Standard font names ImageMagick (used by TextClip) is likely to find in Linux containers DEFAULT_MOVIEPY_FONT = "DejaVu-Sans-Bold" PREFERRED_MOVIEPY_FONT = "Liberation-Sans-Bold" # Often available 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_pil = "DejaVuSans-Bold.ttf" # A more standard Linux font font_paths_to_try = [ self.font_filename_pil, # If in working dir or PATH f"/usr/share/fonts/truetype/dejavu/{self.font_filename_pil}", f"/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf", # Alternative f"/System/Library/Fonts/Supplemental/Arial.ttf", # macOS fallback f"C:/Windows/Fonts/arial.ttf", # Windows fallback f"/usr/local/share/fonts/truetype/mycustomfonts/arial.ttf", # User's previous custom path ] self.font_path_pil_resolved = next( (p for p in font_paths_to_try if os.path.exists(p)), None ) self.font_pil = ImageFont.load_default() # Default self.current_font_size_pil = self.DEFAULT_FONT_SIZE_PIL if self.font_path_pil_resolved: try: self.font_pil = ImageFont.truetype( self.font_path_pil_resolved, self.PREFERRED_FONT_SIZE_PIL ) self.current_font_size_pil = self.PREFERRED_FONT_SIZE_PIL logger.info( f"Pillow font loaded: {self.font_path_pil_resolved} at size {self.current_font_size_pil}." ) # Determine MoviePy font based on loaded PIL font if "dejavu" in self.font_path_pil_resolved.lower(): self.video_overlay_font = "DejaVu-Sans-Bold" elif "liberation" in self.font_path_pil_resolved.lower(): self.video_overlay_font = "Liberation-Sans-Bold" else: # Fallback if custom font doesn't have an obvious ImageMagick name self.video_overlay_font = self.DEFAULT_MOVIEPY_FONT except IOError as e_font_load: logger.error( f"Pillow font loading IOError for '{self.font_path_pil_resolved}': {e_font_load}. Using default." ) else: logger.warning("Custom Pillow font not found. Using default.") 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.elevenlabs_api_key = None self.USE_ELEVENLABS = False 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 self.pexels_api_key = None self.USE_PEXELS = False self.runway_api_key = None self.USE_RUNWAYML = False self.runway_ml_client_instance = None # More specific name # Attempt to initialize Runway client if SDK is present and env var might be set if ( RUNWAYML_SDK_IMPORTED and RunwayMLAPIClient and os.getenv("RUNWAYML_API_SECRET") ): try: self.runway_ml_client_instance = RunwayMLAPIClient() # SDK uses env var self.USE_RUNWAYML = True # Assume enabled if client initializes logger.info( "RunwayML Client initialized from RUNWAYML_API_SECRET env var at startup." ) except Exception as e_runway_init_startup: logger.error( f"Initial RunwayML client init failed (env var RUNWAYML_API_SECRET might be invalid): {e_runway_init_startup}" ) self.USE_RUNWAYML = False logger.info("VisualEngine initialized.") # --- API Key Setters --- def set_openai_api_key(self, api_key): self.openai_api_key = api_key self.USE_AI_IMAGE_GENERATION = bool(api_key) logger.info( f"DALL-E ({self.dalle_model}) status: {'Ready' if self.USE_AI_IMAGE_GENERATION 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 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 status: {'Ready' if self.USE_ELEVENLABS else 'Failed Initialization'} (Using Voice ID: {self.elevenlabs_voice_id})" ) except Exception as e: logger.error( f"ElevenLabs client initialization error: {e}. Service Disabled.", exc_info=True, ) self.USE_ELEVENLABS = False self.elevenlabs_client = None else: self.USE_ELEVENLABS = False logger.info( f"ElevenLabs Service Disabled (API key not provided or SDK import issue)." ) def set_pexels_api_key(self, api_key): self.pexels_api_key = api_key self.USE_PEXELS = bool(api_key) logger.info( f"Pexels Search status: {'Ready' if self.USE_PEXELS else 'Disabled'}" ) def set_runway_api_key(self, api_key): self.runway_api_key = api_key # Store key regardless for potential direct HTTP use if api_key: if RUNWAYML_SDK_IMPORTED and RunwayMLAPIClient: if not self.runway_ml_client_instance: # If not already initialized by env var try: # The RunwayML Python SDK expects the API key via the RUNWAYML_API_SECRET env var. # If it's not set, we set it temporarily for client initialization. original_env_secret = os.getenv("RUNWAYML_API_SECRET") if not original_env_secret: logger.info( "Temporarily setting RUNWAYML_API_SECRET from provided key for SDK client init." ) os.environ["RUNWAYML_API_SECRET"] = api_key self.runway_ml_client_instance = RunwayMLAPIClient() self.USE_RUNWAYML = True # SDK client successfully initialized logger.info( "RunwayML Client initialized successfully using provided API key." ) if not original_env_secret: # Clean up if we set it del os.environ["RUNWAYML_API_SECRET"] logger.info( "Cleared temporary RUNWAYML_API_SECRET env var." ) except Exception as e_client_init: logger.error( f"RunwayML Client initialization via set_runway_api_key failed: {e_client_init}", exc_info=True, ) self.USE_RUNWAYML = False self.runway_ml_client_instance = None else: # Client was already initialized (likely via env var during __init__) self.USE_RUNWAYML = True logger.info( "RunwayML Client was already initialized (likely from env var). API key stored." ) else: # SDK not imported logger.warning( "RunwayML SDK not imported. API key stored, but integration requires SDK. Service effectively disabled." ) self.USE_RUNWAYML = False else: # No API key provided self.USE_RUNWAYML = False self.runway_ml_client_instance = None logger.info("RunwayML Service Disabled (no API key provided).") # --- Helper Methods --- def _image_to_data_uri(self, image_path): try: mime_type, _ = mimetypes.guess_type(image_path) if not mime_type: ext = os.path.splitext(image_path)[1].lower() mime_map = {".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg"} mime_type = mime_map.get(ext, "application/octet-stream") if mime_type == "application/octet-stream": logger.warning( f"Could not determine MIME type for {image_path}, using default." ) with open(image_path, "rb") as image_file: encoded_string = base64.b64encode(image_file.read()).decode("utf-8") data_uri = f"data:{mime_type};base64,{encoded_string}" logger.debug( f"Generated data URI for {os.path.basename(image_path)} (first 100 chars): {data_uri[:100]}..." ) return data_uri except FileNotFoundError: logger.error(f"Image file not found at {image_path} for data URI conversion.") return None except Exception as e: logger.error( f"Error converting image {image_path} to data URI: {e}", exc_info=True ) return None def _map_resolution_to_runway_ratio(self, width, height): ratio_str = f"{width}:{height}" # Gen-4 supports: "1280:720", "720:1280", "1104:832", "832:1104", "960:960", "1584:672" supported_ratios_gen4 = [ "1280:720", "720:1280", "1104:832", "832:1104", "960:960", "1584:672", ] if ratio_str in supported_ratios_gen4: return ratio_str # Fallback or find closest - for now, strict matching or default logger.warning( f"Resolution {ratio_str} not directly in Gen-4 supported list. Defaulting to 1280:720." ) return "1280:720" def _get_text_dimensions(self, text_content, font_object): # (Robust version from before) default_char_height = getattr(font_object, "size", self.current_font_size_pil) if not text_content: return 0, default_char_height try: if hasattr(font_object, "getbbox"): bbox = font_object.getbbox(text_content) w = bbox[2] - bbox[0] h = bbox[3] - bbox[1] return w, h if h > 0 else default_char_height elif hasattr(font_object, "getsize"): w, h = font_object.getsize(text_content) return w, h if h > 0 else default_char_height else: return ( int(len(text_content) * default_char_height * 0.6), int(default_char_height * 1.2), ) except Exception as e: logger.warning(f"Error in _get_text_dimensions: {e}") return ( int(len(text_content) * self.current_font_size_pil * 0.6), int(self.current_font_size_pil * 1.2), ) def _create_placeholder_image_content(self, text_description, filename, size=None): # (Corrected version from previous response) 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 Image)" words = text_description.split() current_line_text = "" for word_idx, word in enumerate(words): prospective_addition = word + (" " if word_idx < len(words) - 1 else "") test_line_text = current_line_text + prospective_addition current_w, _ = self._get_text_dimensions(test_line_text, self.font_pil) if current_w == 0 and test_line_text.strip(): current_w = len(test_line_text) * (self.current_font_size_pil * 0.6) # Estimate if current_w <= max_w: current_line_text = test_line_text else: if current_line_text.strip(): lines.append(current_line_text.strip()) current_line_text = prospective_addition # Start new line if current_line_text.strip(): lines.append(current_line_text.strip()) if not lines and text_description: avg_char_w, _ = self._get_text_dimensions("W", self.font_pil) avg_char_w = avg_char_w or (self.current_font_size_pil * 0.6) chars_per_line = int(max_w / avg_char_w) if avg_char_w > 0 else 20 lines.append( text_description[:chars_per_line] + ("..." if len(text_description) > chars_per_line else "") ) elif not lines: lines.append("(Placeholder Error)") _, single_line_h = self._get_text_dimensions("Ay", self.font_pil) single_line_h = single_line_h if single_line_h > 0 else self.current_font_size_pil + 2 max_lines = ( min(len(lines), (size[1] - (2 * padding)) // (single_line_h + 2)) if single_line_h > 0 else 1 ) max_lines = max(1, max_lines) # Ensure at least one line y_pos = padding + (size[1] - (2 * padding) - max_lines * (single_line_h + 2)) / 2.0 for i in range(max_lines): line_text = lines[i] line_w, _ = self._get_text_dimensions(line_text, self.font_pil) if line_w == 0 and line_text.strip(): line_w = len(line_text) * (self.current_font_size_pil * 0.6) x_pos = (size[0] - line_w) / 2.0 try: d.text((x_pos, y_pos), line_text, font=self.font_pil, fill=(200, 200, 180)) except Exception as e_draw: logger.error(f"Pillow d.text error: {e_draw} for '{line_text}'") y_pos += single_line_h + 2 if i == 6 and max_lines > 7: try: d.text((x_pos, y_pos), "...", font=self.font_pil, fill=(200, 200, 180)) except Exception as e_elip: logger.error(f"Pillow d.text ellipsis error: {e_elip}") break filepath = os.path.join(self.output_dir, filename) try: img.save(filepath) return filepath except Exception as e_save: logger.error( f"Saving placeholder image '{filepath}' error: {e_save}", exc_info=True ) return None def _search_pexels_image(self, query, output_filename_base): # <<< THIS IS THE CORRECTED METHOD >>> 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": "large2x"} base_name_for_pexels, _ = os.path.splitext(output_filename_base) pexels_filename = base_name_for_pexels + f"_pexels_{random.randint(1000,9999)}.jpg" filepath = os.path.join(self.output_dir, pexels_filename) try: logger.info(f"Pexels: Searching 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_details = data["photos"][0] photo_url = photo_details.get("src", {}).get("large2x") if not photo_url: logger.warning( f"Pexels: 'large2x' URL missing for '{effective_query}'. Details: {photo_details}" ) return None image_response = requests.get(photo_url, timeout=60) image_response.raise_for_status() img_data_pil = Image.open(io.BytesIO(image_response.content)) if img_data_pil.mode != "RGB": img_data_pil = img_data_pil.convert("RGB") img_data_pil.save(filepath) logger.info(f"Pexels: Image saved to {filepath}") return filepath else: logger.info(f"Pexels: No photos for '{effective_query}'.") return None except requests.exceptions.RequestException as e_req: logger.error(f"Pexels: RequestException for '{query}': {e_req}", exc_info=False) return None # Less verbose for network except Exception as e: logger.error(f"Pexels: General error for '{query}': {e}", exc_info=True) return None # --- RunwayML Video Generation (Gen-4 Aligned with SDK) --- def _generate_video_clip_with_runwayml( self, text_prompt_for_motion, input_image_path, scene_identifier_filename_base, target_duration_seconds=5, ): if not self.USE_RUNWAYML or not self.runway_ml_client_instance: logger.warning("RunwayML not enabled or client not initialized. Cannot generate video clip.") return None if not input_image_path or not os.path.exists(input_image_path): logger.error( f"Runway Gen-4 requires an input image. Path not provided or invalid: {input_image_path}" ) return None image_data_uri = self._image_to_data_uri(input_image_path) if not image_data_uri: return None runway_duration = 10 if target_duration_seconds >= 8 else 5 # Map to 5s or 10s for Gen-4 runway_ratio_str = self._map_resolution_to_runway_ratio( self.video_frame_size[0], self.video_frame_size[1] ) # Use a more descriptive output filename for Runway videos base_name_for_runway, _ = os.path.splitext(scene_identifier_filename_base) output_video_filename = base_name_for_runway + f"_runway_gen4_d{runway_duration}s.mp4" output_video_filepath = os.path.join(self.output_dir, output_video_filename) logger.info( f"Initiating Runway Gen-4 task: motion='{text_prompt_for_motion[:100]}...', image='{os.path.basename(input_image_path)}', dur={runway_duration}s, ratio='{runway_ratio_str}'" ) try: # Using the RunwayML Python SDK structure task_submission = self.runway_ml_client_instance.image_to_video.create( model="gen4_turbo", prompt_image=image_data_uri, prompt_text=text_prompt_for_motion, # This is the motion prompt duration=runway_duration, ratio=runway_ratio_str, # seed=random.randint(0, 4294967295), # Optional: for reproducibility # Other Gen-4 params (motion_score, upscale, watermark etc. can be added here if available in SDK) ) task_id = task_submission.id logger.info(f"Runway Gen-4 task created with ID: {task_id}. Polling for completion...") poll_interval_seconds = 10 max_polling_duration_seconds = 6 * 60 # 6 minutes start_time = time.time() while time.time() - start_time < max_polling_duration_seconds: time.sleep(poll_interval_seconds) task_details = self.runway_ml_client_instance.tasks.retrieve(id=task_id) logger.info(f"Runway task {task_id} status: {task_details.status}") if task_details.status == "SUCCEEDED": # Determine output URL (this structure might vary based on SDK version) output_url = None if hasattr(task_details, "output") and task_details.output and hasattr( task_details.output, "url" ): output_url = task_details.output.url elif ( hasattr(task_details, "artifacts") and task_details.artifacts and isinstance(task_details.artifacts, list) and len(task_details.artifacts) > 0 ): first_artifact = task_details.artifacts[0] if hasattr(first_artifact, "url"): output_url = first_artifact.url elif hasattr(first_artifact, "download_url"): output_url = first_artifact.download_url if not output_url: logger.error( f"Runway task {task_id} SUCCEEDED, but no output URL found. Details: {vars(task_details) if hasattr(task_details,'__dict__') else str(task_details)}" ) return None logger.info(f"Runway task {task_id} SUCCEEDED. Downloading video from: {output_url}") video_response = requests.get(output_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"Runway Gen-4 video successfully downloaded to: {output_video_filepath}" ) return output_video_filepath elif task_details.status in ["FAILED", "ABORTED", "ERROR"]: # Added ERROR error_msg = ( getattr(task_details, "error_message", None) or getattr(getattr(task_details, "output", None), "error", "Unknown error from Runway task.") ) logger.error( f"Runway task {task_id} final status: {task_details.status}. Error: {error_msg}" ) return None logger.warning( f"Runway task {task_id} timed out polling after {max_polling_duration_seconds} seconds." ) return None except AttributeError as ae: # If SDK methods are not as expected logger.error( f"AttributeError with RunwayML SDK: {ae}. Ensure SDK is up to date and methods/attributes match documentation.", exc_info=True, ) return None except Exception as e_runway_call: logger.error( f"General error during Runway Gen-4 API call or processing: {e_runway_call}", exc_info=True, ) return None def _create_placeholder_video_content(self, text_description, filename, duration=4, size=None): # (Keeping as before) if size is None: size = self.video_frame_size fp = os.path.join(self.output_dir, filename) tc = None try: tc = TextClip( text_description, fontsize=50, color="white", font=self.video_overlay_font, bg_color="black", size=size, method="caption", ).set_duration(duration) tc.write_videofile( fp, fps=24, codec="libx264", preset="ultrafast", logger=None, threads=2 ) logger.info(f"Generic placeholder video: {fp}") return fp except Exception as e: logger.error(f"Generic placeholder video error {fp}: {e}", exc_info=True) return None finally: if tc and hasattr(tc, "close"): tc.close() # --- generate_scene_asset (Main asset generation logic using Runway Gen-4 workflow) --- def generate_scene_asset( self, image_generation_prompt_text, motion_prompt_text_for_video, scene_data, scene_identifier_filename_base, generate_as_video_clip=False, runway_target_duration=5, ): # (Logic updated for improved DALL·E and RunwayML fallback) base_name, _ = os.path.splitext(scene_identifier_filename_base) asset_info = { "path": None, "type": "none", "error": True, "prompt_used": image_generation_prompt_text, "error_message": "Asset generation init failed", } input_image_for_runway_path = None # Use a distinct name for the base image if it's only an intermediate step for video base_image_filename = base_name + ("_base_for_video.png" if generate_as_video_clip else ".png") base_image_filepath = os.path.join(self.output_dir, base_image_filename) # STEP 1: Generate/acquire the base image via DALL·E if self.USE_AI_IMAGE_GENERATION and self.openai_api_key: try: logger.info(f"Calling DALL·E with prompt: {image_generation_prompt_text[:70]}...") response = openai.Image.create( prompt=image_generation_prompt_text, n=1, size=self.image_size_dalle3, model=self.dalle_model, ) image_url = response["data"][0]["url"] ir = requests.get(image_url, timeout=120) ir.raise_for_status() id_img = Image.open(io.BytesIO(ir.content)) if id_img.mode != "RGB": id_img = id_img.convert("RGB") id_img.save(base_image_filepath) logger.info(f"DALL·E base image saved: {base_image_filepath}") input_image_for_runway_path = base_image_filepath asset_info = { "path": base_image_filepath, "type": "image", "error": False, "prompt_used": image_generation_prompt_text, } except openai.error.OpenAIError as e: logger.warning(f"DALL·E error: {e}. Falling back to Pexels or placeholder.") asset_info["error_message"] = str(e) except Exception as e: logger.error(f"Unexpected DALL·E error: {e}", exc_info=True) asset_info["error_message"] = str(e) # STEP 2: If DALL·E failed, try Pexels if asset_info["error"] and self.USE_PEXELS: logger.info("Attempting Pexels fallback for base image.") pqt = scene_data.get( "pexels_search_query_감독", f"{scene_data.get('emotional_beat','')} {scene_data.get('setting_description','')}" ) pp = self._search_pexels_image(pqt, base_image_filename) if pp: input_image_for_runway_path = pp asset_info = { "path": pp, "type": "image", "error": False, "prompt_used": f"Pexels:{pqt}", } else: current_em = asset_info.get("error_message", "") asset_info["error_message"] = (current_em + " Pexels fallback failed.").strip() # STEP 3: If both DALL·E and Pexels failed, create placeholder if asset_info["error"]: logger.warning("Both DALL·E and Pexels failed. Creating placeholder image.") ppt = asset_info.get("prompt_used", image_generation_prompt_text) php = self._create_placeholder_image_content( f"[Placeholder for] {ppt[:70]}...", base_image_filename ) if php: input_image_for_runway_path = php asset_info = { "path": php, "type": "image", "error": False, "prompt_used": ppt, } else: current_em = asset_info.get("error_message", "") asset_info["error_message"] = (current_em + " Placeholder creation failed.").strip() # STEP 4: If a video clip is requested, attempt RunwayML if generate_as_video_clip: if not input_image_for_runway_path or not os.path.exists(input_image_for_runway_path): logger.error("No valid base image for RunwayML. Skipping video generation.") asset_info["error"] = True asset_info["error_message"] = (asset_info.get("error_message", "") + " No base image.").strip() asset_info["type"] = "none" return asset_info if self.USE_RUNWAYML and self.runway_ml_client_instance: video_path = self._generate_video_clip_with_runwayml( motion_prompt_text_for_video, input_image_for_runway_path, base_name, runway_target_duration, ) if video_path and os.path.exists(video_path): asset_info = { "path": video_path, "type": "video", "error": False, "prompt_used": motion_prompt_text_for_video, "base_image_path": input_image_for_runway_path, } else: logger.warning("RunwayML video generation failed. Returning base image instead.") asset_info = { "path": input_image_for_runway_path, "type": "image", "error": True, "prompt_used": image_generation_prompt_text, "error_message": (asset_info.get("error_message", "") + " RunwayML failed.").strip(), } else: logger.warning("RunwayML not enabled or client not initialized. Skipping video generation.") asset_info = { "path": input_image_for_runway_path, "type": "image", "error": True, "prompt_used": image_generation_prompt_text, "error_message": (asset_info.get("error_message", "") + " RunwayML disabled.").strip(), } return asset_info def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"): # (Keep as before - robust enough) if not self.USE_ELEVENLABS or not self.elevenlabs_client or not text_to_narrate: logger.info("ElevenLabs audio skipped.") return None afp = os.path.join(self.output_dir, output_filename) try: logger.info(f"ElevenLabs audio (Voice:{self.elevenlabs_voice_id}): {text_to_narrate[:70]}...") asm = None if hasattr(self.elevenlabs_client, "text_to_speech") and hasattr( self.elevenlabs_client.text_to_speech, "stream" ): asm = self.elevenlabs_client.text_to_speech.stream logger.info("Using ElevenLabs .text_to_speech.stream()") elif hasattr(self.elevenlabs_client, "generate_stream"): asm = self.elevenlabs_client.generate_stream logger.info("Using ElevenLabs .generate_stream()") elif hasattr(self.elevenlabs_client, "generate"): logger.info("Using ElevenLabs .generate()") vp = ( 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) ) ab = self.elevenlabs_client.generate( text=text_to_narrate, voice=vp, model="eleven_multilingual_v2" ) with open(afp, "wb") as f: f.write(ab) logger.info(f"ElevenLabs audio (non-stream) saved: {afp}") return afp else: logger.error("No ElevenLabs audio method available.") return None # If we have a streaming method (asm), use it if asm: vps = {"voice_id": str(self.elevenlabs_voice_id)} if self.elevenlabs_voice_settings: if hasattr(self.elevenlabs_voice_settings, "model_dump"): vps["voice_settings"] = self.elevenlabs_voice_settings.model_dump() elif hasattr(self.elevenlabs_voice_settings, "dict"): vps["voice_settings"] = self.elevenlabs_voice_settings.dict() else: vps["voice_settings"] = self.elevenlabs_voice_settings adi = asm(text=text_to_narrate, model_id="eleven_multilingual_v2", **vps) with open(afp, "wb") as f: for c in adi: if c: f.write(c) logger.info(f"ElevenLabs audio (stream) saved: {afp}") return afp except Exception as e: logger.error(f"ElevenLabs audio error: {e}", exc_info=True) return None # --- assemble_animatic_from_assets (Still contains crucial debug saves for blank video issue) --- def assemble_animatic_from_assets( self, asset_data_list, overall_narration_path=None, output_filename="final_video.mp4", fps=24 ): # (Keep the version with robust image processing, C-contiguous arrays, debug saves, and pix_fmt) if not asset_data_list: logger.warning("No assets for animatic.") return None processed_clips = [] narration_clip = None final_clip = None logger.info(f"Assembling from {len(asset_data_list)} assets. Frame: {self.video_frame_size}.") for i, asset_info in enumerate(asset_data_list): asset_path = asset_info.get("path") asset_type = asset_info.get("type") scene_dur = 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"S{scene_num}: Path='{asset_path}', Type='{asset_type}', Dur='{scene_dur}'s") if not (asset_path and os.path.exists(asset_path)): logger.warning(f"S{scene_num}: Not found '{asset_path}'. Skip.") continue if scene_dur <= 0: logger.warning(f"S{scene_num}: Invalid duration ({scene_dur}s). Skip.") continue current_scene_mvpy_clip = None try: if asset_type == "image": pil_img = Image.open(asset_path) logger.debug(f"S{scene_num}: Loaded img. Mode:{pil_img.mode}, Size:{pil_img.size}") img_rgba = pil_img.convert("RGBA") if pil_img.mode != "RGBA" else pil_img.copy() thumb = img_rgba.copy() rf = Image.Resampling.LANCZOS if hasattr(Image.Resampling, "LANCZOS") else Image.BILINEAR thumb.thumbnail(self.video_frame_size, rf) cv_rgba = Image.new("RGBA", self.video_frame_size, (0, 0, 0, 0)) xo, yo = ( (self.video_frame_size[0] - thumb.width) // 2, (self.video_frame_size[1] - thumb.height) // 2, ) cv_rgba.paste(thumb, (xo, yo), thumb) final_rgb_pil = Image.new("RGB", self.video_frame_size, (0, 0, 0)) final_rgb_pil.paste(cv_rgba, mask=cv_rgba.split()[3]) dbg_path = os.path.join(self.output_dir, f"debug_PRE_NUMPY_S{scene_num}.png") final_rgb_pil.save(dbg_path) logger.info(f"DEBUG: Saved PRE_NUMPY_S{scene_num} to {dbg_path}") frame_np = np.array(final_rgb_pil, dtype=np.uint8) if not frame_np.flags["C_CONTIGUOUS"]: frame_np = np.ascontiguousarray(frame_np, dtype=np.uint8) logger.debug( f"S{scene_num}: NumPy for MoviePy. Shape:{frame_np.shape}, DType:{frame_np.dtype}, C-Contig:{frame_np.flags['C_CONTIGUOUS']}" ) if frame_np.size == 0 or frame_np.ndim != 3 or frame_np.shape[2] != 3: logger.error(f"S{scene_num}: Invalid NumPy. Skip.") continue clip_base = ImageClip(frame_np, transparent=False).set_duration(scene_dur) mvpy_dbg_path = os.path.join(self.output_dir, f"debug_MOVIEPY_FRAME_S{scene_num}.png") clip_base.save_frame(mvpy_dbg_path, t=0.1) logger.info(f"DEBUG: Saved MOVIEPY_FRAME_S{scene_num} to {mvpy_dbg_path}") clip_fx = clip_base try: es = random.uniform(1.03, 1.08) clip_fx = clip_base.fx( vfx.resize, lambda t: 1 + (es - 1) * (t / scene_dur) if scene_dur > 0 else 1 ).set_position("center") except Exception as e: logger.error(f"S{scene_num} Ken Burns error: {e}", exc_info=False) current_scene_mvpy_clip = clip_fx elif asset_type == "video": src_clip = None try: src_clip = VideoFileClip( asset_path, target_resolution=( self.video_frame_size[1], self.video_frame_size[0], ) if self.video_frame_size else None, audio=False, ) tmp_clip = src_clip if src_clip.duration != scene_dur: if src_clip.duration > scene_dur: tmp_clip = src_clip.subclip(0, scene_dur) else: if scene_dur / src_clip.duration > 1.5 and src_clip.duration > 0.1: tmp_clip = src_clip.loop(duration=scene_dur) else: tmp_clip = src_clip.set_duration(src_clip.duration) logger.info( f"S{scene_num} Video clip ({src_clip.duration:.2f}s) shorter than target ({scene_dur:.2f}s)." ) current_scene_mvpy_clip = tmp_clip.set_duration(scene_dur) if current_scene_mvpy_clip.size != list(self.video_frame_size): current_scene_mvpy_clip = current_scene_mvpy_clip.resize(self.video_frame_size) except Exception as e: logger.error(f"S{scene_num} Video load error '{asset_path}':{e}", exc_info=True) continue finally: if src_clip and src_clip is not current_scene_mvpy_clip and hasattr(src_clip, "close"): src_clip.close() else: logger.warning(f"S{scene_num} Unknown asset type '{asset_type}'. Skip.") continue if current_scene_mvpy_clip and key_action: try: to_dur = ( min(current_scene_mvpy_clip.duration - 0.5, current_scene_mvpy_clip.duration * 0.8) if current_scene_mvpy_clip.duration > 0.5 else current_scene_mvpy_clip.duration ) to_start = 0.25 if to_dur > 0: txt_c = 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(to_dur).set_start(to_start).set_position( ("center", 0.92), relative=True ) current_scene_mvpy_clip = CompositeVideoClip( [current_scene_mvpy_clip, txt_c], size=self.video_frame_size, use_bgclip=True ) else: logger.warning(f"S{scene_num}: Text overlay duration is zero. Skip text.") except Exception as e: logger.error(f"S{scene_num} TextClip error:{e}. No text.", exc_info=True) if current_scene_mvpy_clip: processed_clips.append(current_scene_mvpy_clip) logger.info(f"S{scene_num} Processed. Dur:{current_scene_mvpy_clip.duration:.2f}s.") except Exception as e: logger.error(f"MAJOR Error S{scene_num} ({asset_path}):{e}", exc_info=True) finally: if current_scene_mvpy_clip and hasattr(current_scene_mvpy_clip, "close"): try: current_scene_mvpy_clip.close() except: pass if not processed_clips: logger.warning("No clips processed. Abort.") return None td = 0.75 try: logger.info(f"Concatenating {len(processed_clips)} clips.") if len(processed_clips) > 1: final_clip = concatenate_videoclips(processed_clips, padding=-td if td > 0 else 0, method="compose") elif processed_clips: final_clip = processed_clips[0] if not final_clip: logger.error("Concatenation failed.") return None logger.info(f"Concatenated dur:{final_clip.duration:.2f}s") if td > 0 and final_clip.duration > 0: if final_clip.duration > td * 2: final_clip = final_clip.fx(vfx.fadein, td).fx(vfx.fadeout, td) else: final_clip = final_clip.fx(vfx.fadein, min(td, final_clip.duration / 2.0)) if overall_narration_path and os.path.exists(overall_narration_path) and final_clip.duration > 0: try: narration_clip = AudioFileClip(overall_narration_path) final_clip = final_clip.set_audio(narration_clip) logger.info("Narration added.") except Exception as e: logger.error(f"Narration add error:{e}", exc_info=True) elif final_clip.duration <= 0: logger.warning("Video no duration. No audio.") if final_clip and final_clip.duration > 0: op = os.path.join(self.output_dir, output_filename) logger.info(f"Writing video:{op} (Dur:{final_clip.duration:.2f}s)") final_clip.write_videofile( op, 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", ffmpeg_params=["-pix_fmt", "yuv420p"], ) logger.info(f"Video created:{op}") return op else: logger.error("Final clip invalid. No write.") return None except Exception as e: logger.error(f"Video write error:{e}", exc_info=True) return None finally: logger.debug("Closing all MoviePy clips in `assemble_animatic_from_assets` finally block.") all_clips_to_close = processed_clips + ([narration_clip] if narration_clip else []) + ([final_clip] if final_clip else []) for clip_obj_to_close in all_clips_to_close: if clip_obj_to_close and hasattr(clip_obj_to_close, "close"): try: clip_obj_to_close.close() except Exception as e_close: logger.warning( f"Ignoring error while closing a clip: {type(clip_obj_to_close).__name__} - {e_close}" )