# core/visual_engine.py from PIL import Image, ImageDraw, ImageFont, ImageOps from moviepy.editor import (ImageClip, concatenate_videoclips, TextClip, CompositeVideoClip, AudioFileClip) 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 logging logger = logging.getLogger(__name__) logger.setLevel(logging.INFO) # --- ElevenLabs Import --- 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("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.") class VisualEngine: def __init__(self, output_dir="temp_cinegen_media"): 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.video_overlay_font_color = 'white' self.video_overlay_font = 'Liberation-Sans-Bold' # More likely to be found by ImageMagick on Linux 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}.") 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 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 = "Rachel" 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 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 (no API key).'}") def set_elevenlabs_api_key(self,api_key): self.elevenlabs_api_key=api_key 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("ElevenLabs Client Ready.") else: self.USE_ELEVENLABS=False; logger.warning("ElevenLabs client is None post-init.") except Exception as e: logger.error(f"Error initializing ElevenLabs client: {e}. 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 # Logged at import else: logger.info("ElevenLabs API Key not provided. Disabled.") 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).'}") def _get_text_dimensions(self,text_content,font_obj): # No changes if not text_content: return 0,self.font_size_pil try: if hasattr(font_obj,'getbbox'): bbox=font_obj.getbbox(text_content);w=bbox[2]-bbox[0];h=bbox[3]-bbox[1];return w, h if h > 0 else self.font_size_pil elif hasattr(font_obj,'getsize'): w,h=font_obj.getsize(text_content);return w, h if h > 0 else self.font_size_pil else: return int(len(text_content)*self.font_size_pil*0.6),int(self.font_size_pil*1.2 if self.font_size_pil*1.2>0 else self.font_size_pil) except: 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 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+" "; 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()); current_line=word+" " if current_line: lines.append(current_line.strip()) if not lines: lines.append("(Text error or too long for placeholder)") _,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)) y_text=padding + (size[1]-(2*padding) - max_lines_to_display*(single_line_h+2))/2.0 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); 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 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") 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"] 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') img_data.save(filepath); logger.info(f"Pexels image saved: {filepath}"); return filepath else: logger.info(f"No photos found on Pexels for query: '{effective_query}'") 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): # No changes filepath = os.path.join(self.output_dir, scene_identifier_filename) if self.USE_AI_IMAGE_GENERATION and self.openai_api_key: max_retries = 2 for attempt in range(max_retries): try: 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") 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)); 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: 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 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_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) def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"): if not self.USE_ELEVENLABS or not self.elevenlabs_client or not text_to_narrate: logger.info("ElevenLabs conditions not met. Skipping audio generation.") return None audio_filepath = os.path.join(self.output_dir, output_filename) try: logger.info(f"Generating ElevenLabs audio (Voice: {self.elevenlabs_voice_id}) for: {text_to_narrate[:70]}...") # Determine how to pass voice (as object or ID string) voice_param = self.elevenlabs_voice_id if Voice and self.elevenlabs_voice_settings: # Check if Voice & VoiceSettings were imported voice_param = Voice( voice_id=self.elevenlabs_voice_id, # This voice_id must be a valid ID string or a known name settings=self.elevenlabs_voice_settings ) # Use the text_to_speech.stream() method for newer SDK if hasattr(self.elevenlabs_client, 'text_to_speech') and hasattr(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( text=text_to_narrate, voice_id=self.elevenlabs_voice_id, # stream usually takes voice_id string model_id="eleven_multilingual_v2" # or other models like "eleven_monolingual_v1" ) # Fallback to direct .generate() if text_to_speech.stream isn't there (might be an older v1 client or different structure) elif hasattr(self.elevenlabs_client, 'generate'): logger.info("Using elevenlabs_client.generate() as fallback.") audio_data_iterator = self.elevenlabs_client.generate( text=text_to_narrate, voice=voice_param, # This might take the Voice object or just the ID string model="eleven_multilingual_v2" ) else: logger.error("No recognized audio generation method (text_to_speech.stream or generate) 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 except AttributeError as ae: logger.error(f"AttributeError with ElevenLabs client: {ae}. Method or attribute likely missing for installed SDK version.", 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 for video. Target frame size: {self.video_frame_size}") 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 or invalid: {img_path}"); continue try: pil_img = Image.open(img_path) if pil_img.mode != 'RGB': pil_img = pil_img.convert('RGB') # Resize and letterbox/pillarbox img_copy = pil_img.copy() # Use Image.Resampling.LANCZOS for high quality downscaling 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) # This numpy array is correctly sized. # ImageClip should not need to do much resizing now. img_clip = ImageClip(frame_np).set_duration(duration_per_image) end_scale = random.uniform(1.03, 1.08) img_clip = img_clip.fx(vfx.resize, lambda t: 1 + (end_scale - 1) * (t / duration_per_image)) img_clip = img_clip.set_position('center') 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)) # Ensure bg for composite 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 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 overall_narration_path and os.path.exists(overall_narration_path): try: narration_audio_clip = AudioFileClip(overall_narration_path) if narration_audio_clip.duration < final_video_clip_obj.duration: final_video_clip_obj = final_video_clip_obj.subclip(0, narration_audio_clip.duration) final_video_clip_obj = final_video_clip_obj.set_audio(narration_audio_clip) logger.info("Overall narration added to video.") except Exception as e: logger.error(f"Adding overall narration: {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 finally: # Ensure clips are closed for c_item in processed_clips: if hasattr(c_item, 'close'): c_item.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()