# core/visual_engine.py from PIL import Image, ImageDraw, ImageFont 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 from elevenlabs import generate as elevenlabs_generate_audio, set_api_key as elevenlabs_set_api_key_func 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='Arial-Bold' try: self.font = ImageFont.truetype(self.font_path_in_container, self.font_size_pil); print(f"Placeholder font: {self.font_path_in_container}.") except IOError: print(f"Warn: Placeholder font '{self.font_path_in_container}' fail. 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_voice_id = "Rachel" self.pexels_api_key = None; self.USE_PEXELS = False def set_openai_api_key(self,k): self.openai_api_key=k; self.USE_AI_IMAGE_GENERATION=bool(k) print(f"DALL-E ({self.dalle_model}) {'Ready' if k else 'Disabled'}.") def set_elevenlabs_api_key(self,k): self.elevenlabs_api_key=k if k: try: elevenlabs_set_api_key_func(k); self.USE_ELEVENLABS=True; print("ElevenLabs Ready.") except Exception as e: print(f"ElevenLabs key set error: {e}. Disabled."); self.USE_ELEVENLABS=False else: self.USE_ELEVENLABS=False def set_pexels_api_key(self,k): self.pexels_api_key=k; self.USE_PEXELS=bool(k) print(f"Pexels {'Ready' if k else 'Disabled'}.") def _get_text_dimensions(self,t,f): if not t: return 0,self.font_size_pil try: if hasattr(f,'getbbox'): bb=f.getbbox(t);w=bb[2]-bb[0];h=bb[3]-bb[1];return w,h if h>0 else self.font_size_pil elif hasattr(f,'getsize'): w,h=f.getsize(t);return w,h if h>0 else self.font_size_pil else: return int(len(t)*self.font_size_pil*.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(t)*self.font_size_pil*.6),int(self.font_size_pil*1.2) def _create_placeholder_image_content(self,td,fn,s=(1280,720)): img=Image.new('RGB',s,color=(20,20,40));d=ImageDraw.Draw(img);p=25;max_w=s[0]-(2*p);ls=[]; if not td: td="(Placeholder)" ws=td.split();cl="" for w in ws: tl=cl+w+" "; if self._get_text_dimensions(tl,self.font)[0]<=max_w: cl=tl else: if cl:ls.append(cl.strip()) cl=w+" " if cl:ls.append(cl.strip()) if not ls:ls.append("(Text err)") _,sh=self._get_text_dimensions("Ay",self.font);sh=sh if sh>0 else self.font_size_pil+2 max_ls=min(len(ls),(s[1]-2*p)//(sh+2)); yt=p+(s[1]-2*p-max_ls*(sh+2))/2.0 for i in range(max_ls): line=ls[i];lw,_=self._get_text_dimensions(line,self.font);xt=(s[0]-lw)/2.0 d.text((xt,yt),line,font=self.font,fill=(200,200,180));yt+=sh+2 if i==6 and max_ls>7:d.text((xt,yt),"...",font=self.font,fill=(200,200,180));break fp=os.path.join(self.output_dir,fn); try:img.save(fp);return fp except Exception as e:print(f"Err placeholder save: {e}");return None def _search_pexels_image(self, query, output_filename): if not self.USE_PEXELS or not self.pexels_api_key: return None headers = {"Authorization": self.pexels_api_key} params = {"query": query, "per_page": 3, "orientation": "landscape", "size": "large"} pexels_filename = output_filename.replace(".png", f"_pexels_{random.randint(100,999)}.jpg") filepath = os.path.join(self.output_dir, pexels_filename) try: print(f"Searching Pexels for: '{query}'") query_parts = query.split(); effective_query = " ".join(query_parts[:5]) params["query"] = effective_query response = requests.get("https://api.pexels.com/v1/search", headers=headers, params=params, timeout=15) response.raise_for_status(); data = response.json() if data.get("photos"): photo_url = data["photos"][0]["src"]["large2x"] image_response = requests.get(photo_url, timeout=45); 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); print(f"Pexels image saved: {filepath}"); return filepath else: print(f"No photos on Pexels for: '{effective_query}'") except Exception as e: print(f"Pexels error for '{query}': {e}") 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) if self.USE_AI_IMAGE_GENERATION and self.openai_api_key: max_retries = 2 for attempt in range(max_retries): try: print(f"Attempt {attempt+1}: DALL-E ({self.dalle_model}) for: {image_prompt_text[:120]}...") 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: print(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) print(f"AI Image (DALL-E) saved: {filepath}") return filepath except openai.RateLimitError as e: print(f"OpenAI Rate Limit: {e}. Retrying after {5*(attempt+1)}s...") time.sleep(5 * (attempt + 1)) # CORRECTED INDENTATION FOR THIS BLOCK if attempt == max_retries - 1: print("Max retries reached for RateLimitError.") break # Break from the for loop if max retries hit for RateLimitError else: continue # Go to the next attempt in the for loop except openai.APIError as e: print(f"OpenAI API Error: {e}") break # Break from loop, will try Pexels/placeholder except requests.exceptions.RequestException as e: print(f"Requests Error (DALL-E image download): {e}") break # Break from loop except Exception as e: print(f"Generic error (DALL-E gen): {e}") break # Break from loop # This code block is reached if the 'for' loop completes (max retries) # or if it 'break's due to an error other than RateLimitError (where it 'continue's) print("DALL-E generation failed or max retries reached. 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 print("Pexels also failed/disabled. Using placeholder.") return self._create_placeholder_image_content( f"[AI/Pexels Failed] Original Prompt: {image_prompt_text[:100]}...", scene_identifier_filename, size=self.video_frame_size ) else: # AI image generation not enabled return self._create_placeholder_image_content( image_prompt_text, scene_identifier_filename, size=self.video_frame_size ) def generate_narration_audio(self, text_to_narrate, output_filename="narration_overall.mp3"): if not self.USE_ELEVENLABS or not self.elevenlabs_api_key or not text_to_narrate: print("ElevenLabs disabled/no text. Skipping audio."); return None audio_filepath = os.path.join(self.output_dir, output_filename) try: print(f"Generating ElevenLabs audio (Voice: {self.elevenlabs_voice_id}) for: {text_to_narrate[:70]}...") # elevenlabs_set_api_key_func(self.elevenlabs_api_key) # Set key if library requires it per call audio_data = elevenlabs_generate_audio(text=text_to_narrate, voice=self.elevenlabs_voice_id, model="eleven_multilingual_v2") with open(audio_filepath, "wb") as f: f.write(audio_data) print(f"ElevenLabs audio saved: {audio_filepath}"); return audio_filepath except ImportError: print("ElevenLabs library not found. Install it.") except Exception as e: print(f"Error ElevenLabs audio: {e}") 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: return None print(f"Creating video from {len(image_data_list)} image sets.") processed_clips = [] narration_audio_clip = None final_video_clip_obj = None 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)): print(f"Img not found: {img_path}"); continue 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(0,15), random.randint(0,15), random.randint(0,15))) 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) img_clip = ImageClip(frame_np).set_duration(duration_per_image) end_scale = random.uniform(1.05, 1.12) # Ken Burns zoom 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(10,10,20,0.75)', method='caption', align='West', size=(self.video_frame_size[0]*0.9, None), kerning=-1, stroke_color='black', stroke_width=1 ).set_duration(duration_per_image - 1.0).set_start(0.5).set_position(('center', 0.9), relative=True) final_scene_clip = CompositeVideoClip([img_clip, txt_clip], size=self.video_frame_size) else: final_scene_clip = img_clip processed_clips.append(final_scene_clip) except Exception as e: print(f"Error clip for {img_path}: {e}.") if not processed_clips: print("No clips for video."); return None transition = 0.8 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) final_video_clip_obj = final_video_clip_obj.set_audio(narration_audio_clip) if narration_audio_clip.duration < final_video_clip_obj.duration: final_video_clip_obj = final_video_clip_obj.subclip(0, narration_audio_clip.duration) print("Overall narration added.") except Exception as e: print(f"Error adding narration: {e}.") output_path = os.path.join(self.output_dir, output_filename) try: final_video_clip_obj.write_videofile(output_path, fps=fps, codec='libx264', preset='medium', # 'medium' or 'slow' 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") print(f"Video created: {output_path}"); return output_path except Exception as e: print(f"Error writing video: {e}"); return None finally: for c in processed_clips: if hasattr(c, 'close'): c.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()