core/visual_engine.py

# core/visual_engine.py
from PIL import Image, ImageDraw, ImageFont, ImageOps

# --- MONKEY PATCH FOR Image.ANTIALIAS ---
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'):
        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,
    VideoFileClip,
    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 logging

logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)

# --- ElevenLabs Client 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("ElevenLabs client components imported.")
except Exception as e_eleven:
    logger.warning(f"ElevenLabs client import failed: {e_eleven}. Audio disabled.")

# --- RunwayML Client Import (Placeholder) ---
RUNWAYML_SDK_IMPORTED = False
RunwayMLClient = None
try:
    logger.info("RunwayML SDK import is a placeholder.")
except ImportError:
    logger.warning("RunwayML SDK (placeholder) not found. RunwayML disabled.")
except Exception as e_runway_sdk:
    logger.warning(f"Error importing RunwayML SDK (placeholder): {e_runway_sdk}. RunwayML 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 = "DejaVuSans-Bold.ttf"
        font_paths_to_try = [
            self.font_filename,
            "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
            "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf",
            "/System/Library/Fonts/Supplemental/Arial.ttf",
            "C:/Windows/Fonts/arial.ttf",
            "/usr/local/share/fonts/truetype/mycustomfonts/arial.ttf"
        ]
        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 = 'DejaVu-Sans-Bold'

        try:
            if self.font_path_pil:
                self.font = ImageFont.truetype(self.font_path_pil, self.font_size_pil)
                logger.info(f"Pillow font loaded: {self.font_path_pil}.")
            else:
                self.font = ImageFont.load_default()
                logger.warning("Using default Pillow font.")
                self.font_size_pil = 10
        except IOError as e_font:
            logger.error(f"Pillow font loading IOError: {e_font}. Using default.")
            self.font = ImageFont.load_default()
            self.font_size_pil = 10

        self.openai_api_key = None
        self.USE_AI_IMAGE_GENERATION = False
        self.dalle_model = "dall-e-3"
        self.image_size_dalle3 = "1792x1024"
        self.video_frame_size = (1280, 720)

        self.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_client = None

        logger.info("VisualEngine initialized.")

    def set_openai_api_key(self, k):
        self.openai_api_key = k
        self.USE_AI_IMAGE_GENERATION = bool(k)
        logger.info(f"DALL-E ({self.dalle_model}) {'Ready.' if k else 'Disabled.'}")

    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 {'Ready' if self.USE_ELEVENLABS else 'Failed Init'} "
                    f"(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.'}")

    def set_runway_api_key(self, k):
        self.runway_api_key = k
        if k and RUNWAYML_SDK_IMPORTED and RunwayMLClient:
            try:
                self.USE_RUNWAYML = True
                logger.info(
                    f"RunwayML Client (Placeholder SDK) {'Ready.' if self.USE_RUNWAYML else 'Failed Init.'}"
                )
            except Exception as e:
                logger.error(
                    f"RunwayML client (Placeholder SDK) init error: {e}. Disabled.",
                    exc_info=True
                )
                self.USE_RUNWAYML = False
        elif k:
            self.USE_RUNWAYML = True
            logger.info("RunwayML API Key set (direct API or placeholder).")
        else:
            self.USE_RUNWAYML = False
            logger.info("RunwayML Disabled (no API key).")

    def _get_text_dimensions(self, text_content, font_obj):
        default_line_height = getattr(font_obj, 'size', self.font_size_pil)
        if not text_content:
            return 0, default_line_height
        try:
            if hasattr(font_obj, 'getbbox'):
                bbox = font_obj.getbbox(text_content)
                width = bbox[2] - bbox[0]
                height = bbox[3] - bbox[1]
                return width, height if height > 0 else default_line_height
            elif hasattr(font_obj, 'getsize'):
                width, height = font_obj.getsize(text_content)
                return width, height if height > 0 else default_line_height
            else:
                return int(len(text_content) * default_line_height * 0.6), int(default_line_height * 1.2)
        except Exception as e:
            logger.warning(f"Error in _get_text_dimensions for '{text_content[:20]}...': {e}")
            return int(len(text_content) * self.font_size_pil * 0.6), int(self.font_size_pil * 1.2)

    def _create_placeholder_image_content(self, text_description, filename, size=None):
        if size is None:
            size = self.video_frame_size

        img = Image.new('RGB', size, color=(20, 20, 40))
        draw = ImageDraw.Draw(img)
        padding = 25
        max_text_width = size[0] - (2 * padding)
        lines = []

        if not text_description:
            text_description = "(Placeholder: No text description provided)"

        words = text_description.split()
        current_line = ""
        for word in words:
            test_line = current_line + word + " "
            line_width_test, _ = self._get_text_dimensions(test_line.strip(), self.font)
            if line_width_test <= max_text_width:
                current_line = test_line
            else:
                if current_line.strip():
                    lines.append(current_line.strip())
                word_width, _ = self._get_text_dimensions(word, self.font)
                if word_width > max_text_width:
                    avg_char_w = self._get_text_dimensions("A", self.font)[0] or 10
                    chars_that_fit = int(max_text_width / avg_char_w)
                    truncated = (
                        word[:chars_that_fit-3] + "..."
                        if len(word) > chars_that_fit else word
                    )
                    lines.append(truncated)
                    current_line = ""
                else:
                    current_line = word + " "
        if current_line.strip():
            lines.append(current_line.strip())

        if not lines and text_description:
            avg_char_w = self._get_text_dimensions("A", self.font)[0] or 10
            chars_that_fit = int(max_text_width / avg_char_w)
            truncated = (
                text_description[:chars_that_fit-3] + "..."
                if len(text_description) > chars_that_fit else text_description
            )
            lines.append(truncated)
        elif not lines:
            lines.append("(Placeholder Text Error)")

        _, single_line_height = self._get_text_dimensions("Ay", self.font)
        single_line_height = single_line_height if single_line_height > 0 else (self.font_size_pil + 2)
        line_spacing = 2
        max_lines_to_display = min(
            len(lines),
            (size[1] - (2 * padding)) // (single_line_height + line_spacing)
        ) if single_line_height > 0 else 1
        if max_lines_to_display <= 0:
            max_lines_to_display = 1

        total_text_block_height = (
            max_lines_to_display * single_line_height +
            (max_lines_to_display - 1) * line_spacing
        )
        y_text_start = padding + (size[1] - (2 * padding) - total_text_block_height) / 2.0
        current_y = y_text_start

        for i in range(max_lines_to_display):
            line_content = lines[i]
            line_width_actual, _ = self._get_text_dimensions(line_content, self.font)
            x_text = max(padding, (size[0] - line_width_actual) / 2.0)
            draw.text((x_text, current_y), line_content, font=self.font, fill=(200, 200, 180))
            current_y += single_line_height + line_spacing

            if i == 6 and max_lines_to_display > 7 and len(lines) > max_lines_to_display:
                ellipsis_width, _ = self._get_text_dimensions("...", self.font)
                x_ellipsis = max(padding, (size[0] - ellipsis_width) / 2.0)
                draw.text((x_ellipsis, current_y), "...", 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"Error saving placeholder image {filepath}: {e}", exc_info=True)
            return None

    def _search_pexels_image(self, query, output_filename_base):
        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, _ = os.path.splitext(output_filename_base)
        pexels_filename = f"{base_name}_pexels_{random.randint(1000, 9999)}.jpg"
        filepath = os.path.join(self.output_dir, pexels_filename)

        try:
            logger.info(f"Pexels search: '{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["src"]["large2x"]
                logger.info(f"Downloading Pexels image from: {photo_url}")
                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':
                    logger.debug(f"Pexels image mode is {img_data.mode}, converting to RGB.")
                    img_data = img_data.convert('RGB')
                img_data.save(filepath)
                logger.info(f"Pexels image saved successfully: {filepath}")
                return filepath
            else:
                logger.info(f"No photos found on Pexels for query: '{effective_query}'")
                return None
        except requests.exceptions.RequestException as e_req:
            logger.error(f"Pexels request error for query '{query}': {e_req}", exc_info=True)
        except json.JSONDecodeError as e_json:
            logger.error(f"Pexels JSON decode error for query '{query}': {e_json}", exc_info=True)
        except Exception as e:
            logger.error(f"General Pexels error for query '{query}': {e}", exc_info=True)
        return None

    def _generate_video_clip_with_runwayml(self, pt, iip, sifnb, tds=5):
        if not self.USE_RUNWAYML or not self.runway_api_key:
            logger.warning("RunwayML disabled.")
            return None
        if not iip or not os.path.exists(iip):
            logger.error(f"Runway Gen-4 needs input image. Path invalid: {iip}")
            return None

        runway_dur = 10 if tds > 7 else 5
        ovfn = sifnb.replace(".png", f"_runway_gen4_d{runway_dur}s.mp4")
        ovfp = os.path.join(self.output_dir, ovfn)
        logger.info(
            f"Runway Gen-4 (Placeholder) img: {os.path.basename(iip)}, "
            f"motion: '{pt[:100]}...', dur: {runway_dur}s"
        )
        logger.warning("Using PLACEHOLDER video for Runway Gen-4.")

        img_clip = None
        txt_c = None
        final_ph_clip = None
        try:
            img_clip = ImageClip(iip).set_duration(runway_dur)
            txt = (
                f"Runway Gen-4 Placeholder\n"
                f"Input: {os.path.basename(iip)}\n"
                f"Motion: {pt[:50]}..."
            )
            txt_c = TextClip(
                txt,
                fontsize=24,
                color='white',
                font=self.video_overlay_font,
                bg_color='rgba(0,0,0,0.5)',
                size=(int(self.video_frame_size[0] * 0.8), None),
                method='caption'
            ).set_duration(runway_dur).set_position('center')

            final_ph_clip = CompositeVideoClip([img_clip, txt_c], size=img_clip.size)
            final_ph_clip.write_videofile(
                ovfp,
                fps=24,
                codec='libx264',
                preset='ultrafast',
                logger=None,
                threads=2
            )
            logger.info(f"Runway Gen-4 placeholder video: {ovfp}")
            return ovfp
        except Exception as e:
            logger.error(f"Runway Gen-4 placeholder error: {e}", exc_info=True)
            return None
        finally:
            if img_clip and hasattr(img_clip, 'close'):
                img_clip.close()
            if txt_c and hasattr(txt_c, 'close'):
                txt_c.close()
            if final_ph_clip and hasattr(final_ph_clip, 'close'):
                final_ph_clip.close()

    def _create_placeholder_video_content(self, td, fn, dur=4, sz=None):
        if sz is None:
            sz = self.video_frame_size
        fp = os.path.join(self.output_dir, fn)
        tc = None
        try:
            tc = TextClip(
                td,
                fontsize=50,
                color='white',
                font=self.video_overlay_font,
                bg_color='black',
                size=sz,
                method='caption'
            ).set_duration(dur)
            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 error {fp}: {e}", exc_info=True)
            return None
        finally:
            if tc and hasattr(tc, 'close'):
                tc.close()

    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
    ):
        base_name = scene_identifier_filename_base
        asset_info = {
            'path': None,
            'type': 'none',
            'error': True,
            'prompt_used': image_generation_prompt_text,
            'error_message': 'Generation not attempted'
        }
        input_image_for_runway_path = None
        image_filename_for_base = base_name + "_base_image.png"
        temp_image_asset_info = {
            'error': True,
            'prompt_used': image_generation_prompt_text,
            'error_message': 'Base image generation not attempted'
        }

        if self.USE_AI_IMAGE_GENERATION and self.openai_api_key:
            max_r = 2
            for att_n in range(max_r):
                try:
                    img_fp_dalle = os.path.join(self.output_dir, image_filename_for_base)
                    logger.info(
                        f"Attempt {att_n + 1} DALL-E (base img): "
                        f"{image_generation_prompt_text[:100]}..."
                    )
                    cl = openai.OpenAI(api_key=self.openai_api_key, timeout=90.0)
                    r = cl.images.generate(
                        model=self.dalle_model,
                        prompt=image_generation_prompt_text,
                        n=1,
                        size=self.image_size_dalle3,
                        quality="hd",
                        response_format="url",
                        style="vivid"
                    )
                    iu = r.data[0].url
                    rp = getattr(r.data[0], 'revised_prompt', None)
                    if rp:
                        logger.info(f"DALL-E revised: {rp[:100]}...")
                    ir = requests.get(iu, 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(img_fp_dalle)
                    logger.info(f"DALL-E base image: {img_fp_dalle}")
                    input_image_for_runway_path = img_fp_dalle
                    temp_image_asset_info = {
                        'path': img_fp_dalle,
                        'type': 'image',
                        'error': False,
                        'prompt_used': image_generation_prompt_text,
                        'revised_prompt': rp
                    }
                    break
                except openai.RateLimitError as e:
                    logger.warning(f"OpenAI Rate Limit {att_n + 1}: {e}. Retry...")
                    time.sleep(5 * (att_n + 1))
                    temp_image_asset_info['error_message'] = str(e)
                except Exception as e:
                    logger.error(f"DALL-E error: {e}", exc_info=True)
                    temp_image_asset_info['error_message'] = str(e)
                    break

            if temp_image_asset_info['error']:
                logger.warning(f"DALL-E failed after {att_n + 1} attempts for base image.")

        if temp_image_asset_info['error'] and self.USE_PEXELS:
            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, image_filename_for_base)
            if pp:
                input_image_for_runway_path = pp
                temp_image_asset_info = {
                    'path': pp,
                    'type': 'image',
                    'error': False,
                    'prompt_used': f"Pexels: {pqt}"
                }
            else:
                current_em = temp_image_asset_info.get('error_message', "")
                temp_image_asset_info['error_message'] = (current_em + " Pexels failed.").strip()

        if temp_image_asset_info['error']:
            logger.warning("Base image (DALL-E/Pexels) failed. Placeholder base image.")
            ppt = temp_image_asset_info.get('prompt_used', image_generation_prompt_text)
            php = self._create_placeholder_image_content(
                f"[Base Img Placeholder] {ppt[:100]}...",
                image_filename_for_base
            )
            if php:
                input_image_for_runway_path = php
                temp_image_asset_info = {
                    'path': php,
                    'type': 'image',
                    'error': False,
                    'prompt_used': ppt
                }
            else:
                current_em = temp_image_asset_info.get('error_message', "")
                temp_image_asset_info['error_message'] = (current_em + " Base placeholder failed.").strip()

        if generate_as_video_clip:
            if self.USE_RUNWAYML and input_image_for_runway_path:
                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):
                    return {
                        'path': video_path,
                        'type': 'video',
                        'error': False,
                        'prompt_used': motion_prompt_text_for_video,
                        'base_image_path': input_image_for_runway_path
                    }
                else:
                    asset_info = temp_image_asset_info
                    asset_info['error'] = True
                    asset_info['error_message'] = "RunwayML video gen failed; using base image."
                    asset_info['type'] = 'image'
                    return asset_info
            elif not self.USE_RUNWAYML:
                asset_info = temp_image_asset_info
                asset_info['error_message'] = "RunwayML disabled; using base image."
                asset_info['type'] = 'image'
                return asset_info
            else:
                asset_info = temp_image_asset_info
                asset_info['error_message'] = (
                    asset_info.get('error_message', "") +
                    " Base image failed, Runway video not attempted."
                ).strip()
                asset_info['type'] = 'image'
                return asset_info
        else:
            return temp_image_asset_info

    def generate_narration_audio(self, ttn, ofn="narration_overall.mp3"):
        if not self.USE_ELEVENLABS or not self.elevenlabs_client or not ttn:
            logger.info("11L skip.")
            return None

        afp = os.path.join(self.output_dir, ofn)
        try:
            logger.info(f"11L audio (Voice:{self.elevenlabs_voice_id}): {ttn[: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 11L .text_to_speech.stream()")
            elif hasattr(self.elevenlabs_client, 'generate_stream'):
                asm = self.elevenlabs_client.generate_stream
                logger.info("Using 11L .generate_stream()")
            elif hasattr(self.elevenlabs_client, 'generate'):
                logger.info("Using 11L .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=ttn, voice=vp, model="eleven_multilingual_v2")
                with open(afp, "wb") as f:
                    f.write(ab)
                logger.info(f"11L audio (non-stream): {afp}")
                return afp
            else:
                logger.error("No 11L audio method.")
                return None

            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=ttn, model_id="eleven_multilingual_v2", **vps)
            with open(afp, "wb") as f:
                for c in adi:
                    if c:
                        f.write(c)
            logger.info(f"11L audio (stream): {afp}")
            return afp
        except Exception as e:
            logger.error(f"11L audio error: {e}", exc_info=True)
            return None

    def assemble_animatic_from_assets(
        self,
        asset_data_list,
        overall_narration_path=None,
        output_filename="final_video.mp4",
        fps=24
    ):
        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 = (self.video_frame_size[0] - thumb.width) // 2
                    yo = (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. "
                        f"Shape:{frame_np.shape}, DType:{frame_np.dtype}, "
                        f"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) "
                                        f"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
                        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=(int(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
                        )
                    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."
            )
            clips_to_close = (
                processed_clips +
                ([narration_clip] if narration_clip else []) +
                ([final_clip] if final_clip else [])
            )
            for clip_obj in clips_to_close:
                if clip_obj and hasattr(clip_obj, 'close'):
                    try:
                        clip_obj.close()
                    except Exception as e_close:
                        logger.warning(f"Ignoring error while closing a clip: {e_close}")