app.py

import os
import re
import urllib.request
import random
import math
import tempfile
import uuid
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import gradio as gr

# Try to import moviepy; we will fall back gracefully if import fails
try:
    from moviepy.editor import AudioFileClip, VideoFileClip
    from moviepy.audio.fx.all import audio_loop
    HAS_MOVIEPY = True
except ImportError:
    HAS_MOVIEPY = False
    print("Warning: moviepy not installed. Audio mixing will be disabled.")

# Font setup for Korean text rendering
FONT_PATH = "NotoSansKR-Regular.ttf"

def download_font():
    """Downloads a Korean font if it doesn't exist to render subtitles properly."""
    if not os.path.exists(FONT_PATH):
        print("Downloading NotoSansKR-Regular.ttf...")
        url = "https://github.com/google/fonts/raw/main/ofl/notosanskr/NotoSansKR-Regular.ttf"
        try:
            # User-Agent header to avoid block from Github
            req = urllib.request.Request(
                url, 
                headers={'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64)'}
            )
            with urllib.request.urlopen(req) as response, open(FONT_PATH, 'wb') as out_file:
                out_file.write(response.read())
            print("Font download complete.")
        except Exception as e:
            print(f"Error downloading font: {e}")
            # We will use default PIL font if download fails, though Korean won't render.

# Helper function to resize and pad an image to 1080x1920
def fit_image_to_canvas(pil_img, target_w=1080, target_h=1920):
    w, h = pil_img.size
    scale = min(target_w / w, target_h / h)
    new_w = int(w * scale)
    new_h = int(h * scale)
    resized = pil_img.resize((new_w, new_h), Image.Resampling.LANCZOS)
    
    # Create black canvas
    canvas = Image.new("RGB", (target_w, target_h), (0, 0, 0))
    # Paste centered
    x = (target_w - new_w) // 2
    y = (target_h - new_h) // 2
    canvas.paste(resized, (x, y))
    return canvas

# Effect 1: Camera Panning and Zooming on original image
def apply_camera_effect(original_img, effect_name, frame_idx, total_frames, target_w=1080, target_h=1920):
    w, h = original_img.size
    
    # Normalize progress (0.0 to 1.0)
    progress = frame_idx / (total_frames - 1) if total_frames > 1 else 0
    
    if effect_name == "천천히 줌인":
        # Crop from full size down to 85% size
        crop_ratio = 1.0 - 0.15 * progress
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        cx1 = (w - cw) // 2
        cy1 = (h - ch) // 2
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    elif effect_name == "천천히 줌아웃":
        # Crop from 85% size up to full size
        crop_ratio = 0.85 + 0.15 * progress
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        cx1 = (w - cw) // 2
        cy1 = (h - ch) // 2
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    elif effect_name == "좌 -> 우 이동":
        # Crop 85% window, move crop x from left to right
        crop_ratio = 0.85
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        max_shift = w - cw
        cx1 = int(max_shift * progress)
        cy1 = (h - ch) // 2
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    elif effect_name == "우 -> 좌 이동":
        # Crop 85% window, move crop x from right to left
        crop_ratio = 0.85
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        max_shift = w - cw
        cx1 = int(max_shift * (1.0 - progress))
        cy1 = (h - ch) // 2
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    elif effect_name == "위 -> 아래 이동":
        # Crop 85% window, move crop y from top to bottom
        crop_ratio = 0.85
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        max_shift = h - ch
        cx1 = (w - cw) // 2
        cy1 = int(max_shift * progress)
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    elif effect_name == "아래 -> 위 이동":
        # Crop 85% window, move crop y from bottom to top
        crop_ratio = 0.85
        cw = int(w * crop_ratio)
        ch = int(h * crop_ratio)
        max_shift = h - ch
        cx1 = (w - cw) // 2
        cy1 = int(max_shift * (1.0 - progress))
        cropped = original_img.crop((cx1, cy1, cx1 + cw, cy1 + ch))
    else: # Fallback / None
        cropped = original_img
        
    return fit_image_to_canvas(cropped, target_w, target_h)

# Particle class for weather effects
class ParticleSystem:
    def __init__(self, count, p_type, canvas_w=1080, canvas_h=1920):
        self.p_type = p_type
        self.canvas_w = canvas_w
        self.canvas_h = canvas_h
        self.particles = []
        
        for _ in range(count):
            self.particles.append(self._reset_particle(init=True))
            
    def _reset_particle(self, init=False):
        x = random.randint(0, self.canvas_w)
        y = random.randint(0, self.canvas_h) if init else random.randint(-50, 0)
        
        if self.p_type == "눈 내림":
            size = random.uniform(2, 6)
            speed = random.uniform(2, 6)
            drift = random.uniform(-0.5, 0.5)
            color = (255, 255, 255, random.randint(150, 230))
        elif self.p_type == "비 내림":
            size = random.uniform(10, 25) # length
            speed = random.uniform(15, 25)
            drift = 1 # slight wind slant
            color = (180, 200, 255, random.randint(80, 150))
        elif self.p_type == "꽃잎 흩날림":
            size = random.uniform(8, 15)
            speed = random.uniform(3, 7)
            drift = random.uniform(1, 3)
            # Cherry blossom pink tones
            color = (255, random.randint(160, 200), random.randint(180, 220), random.randint(120, 200))
        elif self.p_type == "나뭇잎 흔들림":
            size = random.uniform(10, 18)
            speed = random.uniform(2, 5)
            drift = random.uniform(-1, 2)
            color = (random.randint(80, 130), random.randint(130, 190), random.randint(60, 100), random.randint(120, 200))
        elif self.p_type == "바람 효과":
            size = random.uniform(30, 70)
            speed = random.uniform(1, 3)
            drift = random.uniform(4, 9)
            color = (230, 240, 255, random.randint(90, 150))
        else: # Default/Sunshine
            size = 0
            speed = 0
            drift = 0
            color = (0,0,0,0)
            
        return {"x": x, "y": y, "size": size, "speed": speed, "drift": drift, "color": color, "angle": random.uniform(0, 2*math.pi)}

    def update_and_draw(self, draw, frame_idx):
        if self.p_type == "없음":
            return
            
        if self.p_type == "햇살 반짝임":
            # Draw overlay light leak/shimmer
            overlay = Image.new("RGBA", (self.canvas_w, self.canvas_h), (0,0,0,0))
            overlay_draw = ImageDraw.Draw(overlay)
            
            # Oscillate sunshine center and opacity
            pulse = math.sin(frame_idx * 0.08) * 15 + 60
            center_x = 900 + int(math.sin(frame_idx * 0.05) * 30)
            center_y = 100 + int(math.cos(frame_idx * 0.05) * 20)
            
            # Draw concentric warm soft circles
            for r in range(500, 0, -50):
                alpha = int(pulse * (1 - r / 500) * 0.2)
                overlay_draw.ellipse(
                    (center_x - r, center_y - r, center_x + r, center_y + r),
                    fill=(255, 245, 200, alpha)
                )
            draw.bitmap((0, 0), overlay.convert("L"), fill=(255, 245, 200)) # Simple paste helper or blend
            return

        if self.p_type == "구름 천천히 이동":
            shift = int((frame_idx * 0.6) % (self.canvas_w + 400)) - 300
            for base_x, base_y, scale in [(shift, 180, 1.0), (shift + 430, 320, 0.75), (shift - 520, 520, 0.9)]:
                color = (225, 232, 242)
                draw.ellipse((base_x, base_y, base_x + int(260 * scale), base_y + int(90 * scale)), fill=color)
                draw.ellipse((base_x + int(90 * scale), base_y - int(35 * scale), base_x + int(340 * scale), base_y + int(95 * scale)), fill=color)
            return
            
        # Standard particle types
        for p in self.particles:
            p["y"] += p["speed"]
            p["x"] += p["drift"]
            
            if self.p_type in ["꽃잎 흩날림", "나뭇잎 흔들림"]:
                p["x"] += math.sin(frame_idx * 0.1 + p["angle"]) * 2 # Swaying path
                
            # If off screen, reset
            if p["y"] > self.canvas_h or p["x"] < -20 or p["x"] > self.canvas_w + 20:
                p.update(self._reset_particle(init=False))
                
            # Draw particle
            x, y = p["x"], p["y"]
            sz = p["size"]
            col = p["color"]
            
            if self.p_type == "눈 내림":
                draw.ellipse((x - sz, y - sz, x + sz, y + sz), fill=col[:3])
            elif self.p_type == "비 내림":
                draw.line((x, y, x + p["drift"] * 3, y + sz), fill=col[:3], width=1)
            elif self.p_type == "꽃잎 흩날림":
                p["angle"] += 0.05
                draw.ellipse((x - sz//2, y - sz//3, x + sz//2, y + sz//3), fill=col[:3])
            elif self.p_type == "나뭇잎 흔들림":
                p["angle"] += 0.04
                draw.ellipse((x - sz//2, y - sz//3, x + sz//2, y + sz//3), fill=col[:3])
                draw.line((x, y, x + sz//2, y + sz//3), fill=(70, 110, 55), width=1)
            elif self.p_type == "바람 효과":
                draw.arc((x, y, x + sz, y + sz // 2), start=190, end=350, fill=col[:3], width=2)

# Helper function to wrap text for subtitles
def wrap_text(text, font, max_width):
    lines = []
    words = list(text) # In Korean, wrap by character rather than space is often preferred if long, but we can do space or chars.
    
    # We will split by spaces first
    space_words = text.split(' ')
    current_line = ""
    
    for word in space_words:
        test_line = current_line + " " + word if current_line else word
        # Calculate width
        w = font.getbbox(test_line)[2]
        if w <= max_width:
            current_line = test_line
        else:
            if current_line:
                lines.append(current_line)
            current_line = word
            
    if current_line:
        lines.append(current_line)
    return lines

# Function to draw subtitles with custom styles
def draw_subtitles(canvas, text, font_size_style, font_color_style, use_bg):
    draw = ImageDraw.Draw(canvas, "RGBA")
    
    # Choose font size
    size = 46 if font_size_style == "큰 글씨" else 32
    
    try:
        download_font()
        font = ImageFont.truetype(FONT_PATH, size)
    except IOError:
        font = ImageFont.load_default()
        
    color = (255, 255, 255) if font_color_style == "흰색" else (255, 242, 0)
    
    # Wrap text
    max_w = 960 # 1080 - 120 padding
    wrapped_lines = wrap_text(text, font, max_w)
    
    # Calculate height and positions
    line_spacing = 8
    total_height = sum([font.getbbox(line)[3] - font.getbbox(line)[1] for line in wrapped_lines]) + (len(wrapped_lines) - 1) * line_spacing
    
    start_y = 1750 - total_height // 2
    
    # Draw background box if enabled
    if use_bg and wrapped_lines:
        bg_padding = 15
        box_y1 = start_y - bg_padding
        box_y2 = start_y + total_height + bg_padding
        
        # Transparent background box
        # Create a temp transparent layer to blend alpha box
        overlay = Image.new("RGBA", canvas.size, (0, 0, 0, 0))
        overlay_draw = ImageDraw.Draw(overlay)
        overlay_draw.rectangle((60, box_y1, 1020, box_y2), fill=(0, 0, 0, 140))
        canvas.paste(overlay, (0, 0), overlay)
        
    # Draw text lines centered
    current_y = start_y
    for line in wrapped_lines:
        w = font.getbbox(line)[2]
        x = (1080 - w) // 2
        draw.text((x, current_y), line, fill=color, font=font)
        current_y += (font.getbbox(line)[3] - font.getbbox(line)[1]) + line_spacing

def generate_default_subtitles(video_type, description, num_images):
    desc = description if description else "프로그램 활동"
    templates = {
        "기관홍보형": [
            f"오늘도 활기차게 시작하는 {desc}",
            "서로 돕고 웃음꽃 피는 따뜻한 돌봄 현장",
            "어르신들의 안전하고 든든한 보금자리",
            "가족의 마음으로 함께 동행하겠습니다",
            "늘 곁에서 사랑과 정성으로 보살핍니다"
        ],
        "감동다큐형": [
            "주름진 손끝에서 피어나는 이야기",
            "함께 나누는 미소 속에 흐르는 따뜻한 시간",
            "그 시절 그 고운 마음 그대로",
            "어르신의 소중한 인생 여정을 응원합니다",
            "기억 속에 담긴 따뜻한 추억 한 조각"
        ],
        "생신축하형": [
            "오늘 세상에서 가장 특별하고 기쁜 날",
            "어르신의 소중한 생신을 온 마음으로 축하드립니다",
            "늘 건강하시고 꽃길만 걸으소서",
            "함께해주셔서 참 고맙고 행복합니다",
            "언제나 건강하고 평온하시기를 기도합니다"
        ],
        "창작활동형": [
            "손끝의 정성으로 빚어내는 창작 활동",
            "집중하시는 눈빛 속 피어나는 열정",
            "어르신들의 창의력이 가득 담긴 귀한 작품",
            "만족스러운 미소와 성취감의 순간",
            "자신감과 행복이 자라나는 미술 교실"
        ],
        "작품전시형": [
            "세상에 단 하나뿐인 고귀한 예술 작품",
            "한 땀 한 땀 흘린 정성과 손길",
            "마음의 색깔로 그려낸 어르신의 우주",
            "모두가 감탄한 오늘의 명작 전시회",
            "예술로 소통하고 치유받는 특별한 시간"
        ],
        "추억회상형": [
            "아련히 떠오르는 그 시절 우리들의 추억",
            "사진첩 속에 깃든 그리운 이야기",
            "가슴속 깊이 묻어둔 고운 기억들",
            "다시금 꺼내어 웃음 지어보는 날",
            "흘러간 세월마저 아름다운 인생의 조각"
        ],
        "자서전형": [
            "빛나는 청춘을 지나 도달한 아름다운 계절",
            "지혜와 사랑으로 일구어온 삶의 궤적",
            "가족들의 든든한 뿌리가 되어주신 고마운 분",
            "어르신의 삶 자체가 가장 위대한 소설입니다",
            "따뜻한 발자취를 온 마음으로 기록합니다"
        ]
    }
    
    sub_list = templates.get(video_type, templates["감동다큐형"])
    
    # If the user put description, let's format it dynamically
    result = []
    for i in range(num_images):
        idx = i % len(sub_list)
        result.append(sub_list[idx])
    return result

def get_file_path(file_value):
    if isinstance(file_value, str):
        return file_value
    if hasattr(file_value, "path"):
        return file_value.path
    if isinstance(file_value, dict):
        return file_value.get("path") or file_value.get("name")
    return file_value

# Core generator function for Gradio interface
def make_image_drama_video(image_files, video_type, camera_effect, natural_effect,
                           activity_desc, subtitle_mode, subtitle_input, font_size, font_color, use_bg, music_file):
    
    if not image_files or len(image_files) < 3:
        return None, "최소 3장의 사진을 업로드해 주세요. v1은 3~5장을 권장합니다."
        
    if len(image_files) > 5:
        image_files = image_files[:5]
        
    num_images = len(image_files)
    
    # Step 1: Handle subtitles. Institutions either enter subtitles directly or create a clean video without text.
    use_subtitles = subtitle_mode == "직접입력"
    subtitles = []
    if use_subtitles:
        if not subtitle_input or not subtitle_input.strip():
            return None, "자막 방식을 직접입력으로 선택한 경우, 자막 내용을 입력해 주세요."
        subtitles = [s.strip() for s in re.split(r"[,\n]+", subtitle_input) if s.strip()]
        if len(subtitles) < num_images:
            subtitles += [""] * (num_images - len(subtitles))
        subtitles = subtitles[:num_images]
        
    # Step 2: Render Frames
    fps = 24
    frames_per_image = 96  # 4 seconds per image at 24fps
    total_frames = num_images * frames_per_image
    
    # Initialize particle system for natural effects
    p_count = 0
    if natural_effect in ["눈 내림", "비 내림"]:
        p_count = 50
    elif natural_effect == "꽃잎 흩날림":
        p_count = 35
    elif natural_effect == "나뭇잎 흔들림":
        p_count = 30
    elif natural_effect == "바람 효과":
        p_count = 18
    particles = ParticleSystem(p_count, natural_effect)
    
    import cv2

    run_id = uuid.uuid4().hex
    output_video_path = os.path.join(tempfile.gettempdir(), f"care_cinema_studio_image_drama_{run_id}.mp4")
    silent_video_path = os.path.join(tempfile.gettempdir(), f"care_cinema_studio_image_drama_silent_{run_id}.mp4")
    for path in [output_video_path, silent_video_path]:
        if os.path.exists(path):
            try:
                os.remove(path)
            except Exception:
                pass

    fourcc = cv2.VideoWriter_fourcc(*"mp4v")
    video_writer = cv2.VideoWriter(silent_video_path, fourcc, fps, (1080, 1920))
    if not video_writer.isOpened():
        return None, "영상 파일을 준비하는 중 오류가 발생했습니다. 잠시 후 다시 시도해 주세요."

    for i, img_file in enumerate(image_files):
        # Open and load image
        img_path = get_file_path(img_file)
        try:
            original_img = Image.open(img_path).convert("RGB")
        except Exception as e:
            print(f"Error opening image {img_path}: {e}")
            original_img = Image.new("RGB", (1080, 1920), (50, 50, 50))
            
        current_subtitle = subtitles[i] if use_subtitles else ""
        
        # Select current camera effect
        eff = camera_effect
        if camera_effect == "랜덤 시네마 효과":
            random.seed(i) # Fixed random per photo segment
            eff = random.choice([
                "천천히 줌인", "천천히 줌아웃", "좌 -> 우 이동", 
                "우 -> 좌 이동", "위 -> 아래 이동", "아래 -> 위 이동"
            ])
            
        for f in range(frames_per_image):
            # Apply camera motion
            frame = apply_camera_effect(original_img, eff, f, frames_per_image)
            
            # Apply natural overlay effects
            frame_draw = ImageDraw.Draw(frame)
            global_frame_idx = i * frames_per_image + f
            particles.update_and_draw(frame_draw, global_frame_idx)
            
            if use_subtitles and current_subtitle:
                draw_subtitles(frame, current_subtitle, font_size, font_color, use_bg)

            video_writer.write(cv2.cvtColor(np.array(frame), cv2.COLOR_RGB2BGR))

    video_writer.release()

    # Step 3: Add optional uploaded music. Silent output remains the safe default.
    music_path = get_file_path(music_file) if music_file else None
    if music_path and HAS_MOVIEPY:
        clip = None
        audio_clip = None
        try:
            clip = VideoFileClip(silent_video_path)
            audio_clip = AudioFileClip(music_path)
            if audio_clip.duration < clip.duration:
                audio_clip = audio_loop(audio_clip, duration=clip.duration)
            else:
                audio_clip = audio_clip.subclip(0, clip.duration)
            clip = clip.set_audio(audio_clip)
            clip.write_videofile(
                output_video_path,
                codec="libx264",
                audio_codec="aac",
                temp_audiofile=os.path.join(tempfile.gettempdir(), f"ccs-temp-audio-{run_id}.m4a"),
                remove_temp=True,
                fps=fps,
            )
            if os.path.exists(silent_video_path):
                os.remove(silent_video_path)
        except Exception as ve:
            print(f"Audio mixing failed, returning silent video: {ve}")
            os.replace(silent_video_path, output_video_path)
        finally:
            if audio_clip is not None:
                audio_clip.close()
            if clip is not None:
                clip.close()
    else:
        os.replace(silent_video_path, output_video_path)
            
    status = (
        "영상 생성이 완료되었습니다. "
        f"{num_images}장의 사진으로 9:16 쇼츠 형식의 Image Drama 영상을 만들었습니다. "
        f"자막 방식: {subtitle_mode}"
    )
    return output_video_path, status

# CSS for Care Cinema Studio.
custom_css = """
.gradio-container {
    background-color: #0b111e;
    color: #e5e7eb;
    font-family: 'Outfit', 'Inter', -apple-system, sans-serif;
}
.header-box {
    background: linear-gradient(135deg, #3b82f6 0%, #8b5cf6 50%, #d946ef 100%);
    padding: 2.5rem;
    border-radius: 8px;
    text-align: center;
    margin-bottom: 2rem;
    box-shadow: 0 10px 30px -5px rgba(139, 92, 246, 0.4);
    border: 1px solid rgba(255, 255, 255, 0.1);
}
.header-box h1 {
    color: #ffffff;
    font-size: 2.8rem;
    font-weight: 800;
    margin-bottom: 0.5rem;
    letter-spacing: -0.03em;
    text-shadow: 0 2px 10px rgba(0, 0, 0, 0.3);
}
.header-box p {
    color: #e0e7ff;
    font-size: 1.2rem;
    font-weight: 300;
}
.primary-btn {
    background: linear-gradient(90deg, #3b82f6 0%, #8b5cf6 100%) !important;
    border: none !important;
    color: white !important;
    font-weight: 600 !important;
    transition: all 0.3s cubic-bezier(0.4, 0, 0.2, 1) !important;
    border-radius: 8px !important;
    box-shadow: 0 4px 12px rgba(139, 92, 246, 0.2) !important;
}
.primary-btn:hover {
    transform: translateY(-2px);
    box-shadow: 0 8px 25px rgba(139, 92, 246, 0.45) !important;
}
.glass-panel {
    background: rgba(17, 24, 39, 0.65) !important;
    backdrop-filter: blur(12px) !important;
    border: 1px solid rgba(255, 255, 255, 0.08) !important;
    border-radius: 8px !important;
    padding: 1.5rem !important;
}
"""

# Build Gradio UI
with gr.Blocks(
    title="Care Cinema Studio - Image Drama Generator",
    css=custom_css,
    theme=gr.themes.Soft(primary_hue="violet", secondary_hue="blue"),
) as demo:
    
    # Top Banner Header
    gr.HTML(
        """
        <div class="header-box">
            <h1>🎬 Care Cinema Studio (CCS)</h1>
            <p>Turn Photos Into Stories | 사진을 이야기로 만드는 AI 영상 생성기</p>
        </div>
        """
    )
    
    with gr.Row():
        with gr.Column(scale=1, elem_classes="glass-panel"):
            gr.Markdown("### 📸 사진 및 비디오 설정")
            gr.Markdown("무료 v1 기준: 하루 2회 생성, 사진 3~5장, 최대 20초 MP4 쇼츠")
            image_input = gr.File(
                file_count="multiple", 
                file_types=["image"], 
                type="filepath",
                label="사진 업로드 (3~5장 권장)"
            )
            
            video_type_dropdown = gr.Dropdown(
                choices=["기관홍보형", "감동다큐형", "생신축하형", "창작활동형", "작품전시형", "추억회상형"], 
                value="창작활동형", 
                label="영상 유형"
            )
            
            activity_desc_input = gr.Textbox(
                label="활동 상세 설명 (AI 자막 생성을 돕습니다)",
                placeholder="예: 어르신들의 인지 발달 및 손가락 운동을 위한 클레이 아트 물고기 만들기 미술 치료 프로그램 진행",
                value="어르신들의 소근육 자극과 성취감 획득을 위한 색찰흙 창작 활동",
                lines=2
            )
            
            gr.Markdown("### 🎨 시네마틱 효과")
            camera_effect_dropdown = gr.Dropdown(
                choices=["천천히 줌인", "천천히 줌아웃", "좌 -> 우 이동", "우 -> 좌 이동", "위 -> 아래 이동", "아래 -> 위 이동", "랜덤 시네마 효과"], 
                value="천천히 줌인", 
                label="카메라 효과"
            )
            
            natural_effect_dropdown = gr.Dropdown(
                choices=["없음", "구름 천천히 이동", "바람 효과", "나뭇잎 흔들림", "꽃잎 흩날림", "햇살 반짝임", "비 내림", "눈 내림"], 
                value="없음", 
                label="자연 효과 (오버레이)"
            )
            
            gr.Markdown("### 💬 자막")
            subtitle_mode_radio = gr.Radio(
                choices=["직접입력", "자막없음"],
                value="직접입력",
                label="자막 방식"
            )
            subtitle_input_field = gr.Textbox(
                label="기관 입력 자막 (쉼표 또는 줄바꿈 구분)",
                placeholder="예: 오늘도 즐거운 미술 시간\n손끝에서 피어난 예쁜 작품\n함께 웃는 따뜻한 하루"
            )
            
            with gr.Row():
                font_size_radio = gr.Radio(choices=["큰 글씨", "작은 글씨"], value="큰 글씨", label="글자 크기")
                font_color_radio = gr.Radio(choices=["흰색", "노란색"], value="흰색", label="글자 색상")
                
            use_bg_checkbox = gr.Checkbox(label="검정 반투명 자막 배경 적용", value=True)
            
            gr.Markdown("### 🎵 사운드")
            audio_input = gr.Audio(
                sources=["upload"],
                type="filepath", 
                label="배경 음악 (MP3 업로드, 선택)"
            )
            
            submit_btn = gr.Button("Image Drama 영상 생성", elem_classes="primary-btn")
            
        with gr.Column(scale=1, elem_classes="glass-panel"):
            gr.Markdown("### 🎥 생성된 시네마틱 쇼츠 (9:16)")
            video_output = gr.Video(label="완료된 영상")
            
            gr.Markdown("### 생성 상태")
            status_output = gr.Markdown(value="사진을 업로드하고 생성을 시작하면 이곳에 완료 안내가 표시됩니다.")
            
    # Connect trigger
    submit_btn.click(
        fn=make_image_drama_video,
        inputs=[
            image_input,
            video_type_dropdown,
            camera_effect_dropdown,
            natural_effect_dropdown,
            activity_desc_input,
            subtitle_mode_radio,
            subtitle_input_field,
            font_size_radio,
            font_color_radio,
            use_bg_checkbox,
            audio_input
        ],
        outputs=[video_output, status_output]
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)