app.py

import spaces
import logging
from datetime import datetime
from pathlib import Path
import gradio as gr
import torch
import torchaudio
import os
import requests
from transformers import pipeline
import tempfile
import numpy as np
from einops import rearrange
import cv2
from scipy.io import wavfile
import librosa
import json
from typing import Optional, Tuple, List
import atexit

# 환경 변수 설정으로 torch.load 체크 우회 (임시 해결책)
os.environ["TRANSFORMERS_ALLOW_UNSAFE_DESERIALIZATION"] = "1"

try:
    import mmaudio
except ImportError:
    os.system("pip install -e .")
    import mmaudio

from mmaudio.eval_utils import (ModelConfig, all_model_cfg, generate, load_video, make_video,
                                setup_eval_logging)
from mmaudio.model.flow_matching import FlowMatching
from mmaudio.model.networks import MMAudio, get_my_mmaudio
from mmaudio.model.sequence_config import SequenceConfig
from mmaudio.model.utils.features_utils import FeaturesUtils

# 로깅 설정
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
log = logging.getLogger()

# CUDA 설정
if torch.cuda.is_available():
    device = torch.device("cuda")
    torch.backends.cuda.matmul.allow_tf32 = True
    torch.backends.cudnn.allow_tf32 = True
    torch.backends.cudnn.benchmark = True
else:
    device = torch.device("cpu")

dtype = torch.bfloat16

# 모델 설정
model: ModelConfig = all_model_cfg['large_44k_v2']
model.download_if_needed()
output_dir = Path('./output/gradio')

setup_eval_logging()

# 번역기 설정
try:
    translator = pipeline("translation", 
                         model="Helsinki-NLP/opus-mt-ko-en", 
                         device="cpu",
                         use_fast=True,
                         trust_remote_code=False)
except Exception as e:
    log.warning(f"Failed to load translation model with safetensors: {e}")
    try:
        translator = pipeline("translation", 
                             model="Helsinki-NLP/opus-mt-ko-en", 
                             device="cpu")
    except Exception as e2:
        log.error(f"Failed to load translation model: {e2}")
        translator = None

PIXABAY_API_KEY = "33492762-a28a596ec4f286f84cd328b17"

def cleanup_temp_files():
    temp_dir = tempfile.gettempdir()
    for file in os.listdir(temp_dir):
        if file.endswith(('.mp4', '.flac')):
            try:
                os.remove(os.path.join(temp_dir, file))
            except:
                pass

atexit.register(cleanup_temp_files)

def get_model() -> tuple[MMAudio, FeaturesUtils, SequenceConfig]:
    with torch.cuda.device(device):
        seq_cfg = model.seq_cfg
        net: MMAudio = get_my_mmaudio(model.model_name).to(device, dtype).eval()
        net.load_weights(torch.load(model.model_path, map_location=device, weights_only=True))
        log.info(f'Loaded weights from {model.model_path}')

        feature_utils = FeaturesUtils(
            tod_vae_ckpt=model.vae_path,
            synchformer_ckpt=model.synchformer_ckpt,
            enable_conditions=True,
            mode=model.mode,
            bigvgan_vocoder_ckpt=model.bigvgan_16k_path,
            need_vae_encoder=False
        ).to(device, dtype).eval()

        return net, feature_utils, seq_cfg

net, feature_utils, seq_cfg = get_model()

def translate_prompt(text):
    try:
        if translator is None:
            return text
            
        if text and any(ord(char) >= 0x3131 and ord(char) <= 0xD7A3 for char in text):
            with torch.no_grad():
                translation = translator(text)[0]['translation_text']
            return translation
        return text
    except Exception as e:
        logging.error(f"Translation error: {e}")
        return text

@torch.no_grad()
def search_videos(query):
    try:
        query = translate_prompt(query)
        return search_pixabay_videos(query, PIXABAY_API_KEY)
    except Exception as e:
        logging.error(f"Video search error: {e}")
        return []

def search_pixabay_videos(query, api_key):
    try:
        base_url = "https://pixabay.com/api/videos/"
        params = {
            "key": api_key,
            "q": query,
            "per_page": 40
        }
        
        response = requests.get(base_url, params=params)
        if response.status_code == 200:
            data = response.json()
            return [video['videos']['large']['url'] for video in data.get('hits', [])]
        return []
    except Exception as e:
        logging.error(f"Pixabay API error: {e}")
        return []

@spaces.GPU
@torch.inference_mode()
def video_to_audio(video: gr.Video, prompt: str, negative_prompt: str, seed: int, num_steps: int,
                   cfg_strength: float, duration: float):
    prompt = translate_prompt(prompt)
    negative_prompt = translate_prompt(negative_prompt)

    rng = torch.Generator(device=device)
    rng.manual_seed(seed)
    fm = FlowMatching(min_sigma=0, inference_mode='euler', num_steps=num_steps)

    clip_frames, sync_frames, duration = load_video(video, duration)
    clip_frames = clip_frames.unsqueeze(0)
    sync_frames = sync_frames.unsqueeze(0)
    seq_cfg.duration = duration
    net.update_seq_lengths(seq_cfg.latent_seq_len, seq_cfg.clip_seq_len, seq_cfg.sync_seq_len)

    audios = generate(clip_frames,
                      sync_frames, [prompt],
                      negative_text=[negative_prompt],
                      feature_utils=feature_utils,
                      net=net,
                      fm=fm,
                      rng=rng,
                      cfg_strength=cfg_strength)
    audio = audios.float().cpu()[0]

    video_save_path = tempfile.NamedTemporaryFile(delete=False, suffix='.mp4').name
    make_video(video,
               video_save_path,
               audio,
               sampling_rate=seq_cfg.sampling_rate,
               duration_sec=seq_cfg.duration)
    
    # 정보 로그 생성
    info_log = f"""✅ VIDEO TO AUDIO COMPLETE!
{'=' * 50}
🎬 Video Info:
   • Duration: {duration:.2f} seconds
{'=' * 50}
⚙️ Generation Settings:
   • Seed: {seed}
   • Steps: {num_steps}
   • CFG Strength: {cfg_strength}
{'=' * 50}
📝 Prompts:
   • Prompt: {prompt[:40]}{'...' if len(prompt) > 40 else ''}
   • Negative: {negative_prompt[:30]}{'...' if len(negative_prompt) > 30 else ''}
{'=' * 50}
💾 Video with audio ready!"""
    
    return video_save_path, info_log

@spaces.GPU
@torch.inference_mode()
def text_to_audio(prompt: str, negative_prompt: str, seed: int, num_steps: int, cfg_strength: float,
                  duration: float):
    prompt = translate_prompt(prompt)
    negative_prompt = translate_prompt(negative_prompt)

    rng = torch.Generator(device=device)
    rng.manual_seed(seed)
    fm = FlowMatching(min_sigma=0, inference_mode='euler', num_steps=num_steps)

    clip_frames = sync_frames = None
    seq_cfg.duration = duration
    net.update_seq_lengths(seq_cfg.latent_seq_len, seq_cfg.clip_seq_len, seq_cfg.sync_seq_len)

    audios = generate(clip_frames,
                      sync_frames, [prompt],
                      negative_text=[negative_prompt],
                      feature_utils=feature_utils,
                      net=net,
                      fm=fm,
                      rng=rng,
                      cfg_strength=cfg_strength)
    audio = audios.float().cpu()[0]

    audio_save_path = tempfile.NamedTemporaryFile(delete=False, suffix='.flac').name
    torchaudio.save(audio_save_path, audio, seq_cfg.sampling_rate)
    
    # 정보 로그 생성
    info_log = f"""✅ TEXT TO AUDIO COMPLETE!
{'=' * 50}
🎵 Audio Info:
   • Duration: {duration:.2f} seconds
   • Sample Rate: {seq_cfg.sampling_rate} Hz
{'=' * 50}
⚙️ Generation Settings:
   • Seed: {seed}
   • Steps: {num_steps}
   • CFG Strength: {cfg_strength}
{'=' * 50}
📝 Prompts:
   • Prompt: {prompt[:40]}{'...' if len(prompt) > 40 else ''}
   • Negative: {negative_prompt[:30]}{'...' if len(negative_prompt) > 30 else ''}
{'=' * 50}
💾 Audio ready to download!"""
    
    return audio_save_path, info_log


# ============================================
# 🎨 Comic Classic Theme - Toon Playground
# ============================================

css = """
/* ===== 🎨 Google Fonts Import ===== */
@import url('https://fonts.googleapis.com/css2?family=Bangers&family=Comic+Neue:wght@400;700&display=swap');

/* ===== 🎨 Comic Classic 배경 - 빈티지 페이퍼 + 도트 패턴 ===== */
.gradio-container {
    background-color: #FEF9C3 !important;
    background-image: 
        radial-gradient(#1F2937 1px, transparent 1px) !important;
    background-size: 20px 20px !important;
    min-height: 100vh !important;
    font-family: 'Comic Neue', cursive, sans-serif !important;
}

/* ===== 허깅페이스 상단 요소 숨김 ===== */
.huggingface-space-header,
#space-header,
.space-header,
[class*="space-header"],
.svelte-1ed2p3z,
.space-header-badge,
.header-badge,
[data-testid="space-header"],
.svelte-kqij2n,
.svelte-1ax1toq,
.embed-container > div:first-child {
    display: none !important;
    visibility: hidden !important;
    height: 0 !important;
    width: 0 !important;
    overflow: hidden !important;
    opacity: 0 !important;
    pointer-events: none !important;
}

/* ===== Footer 완전 숨김 ===== */
footer,
.footer,
.gradio-container footer,
.built-with,
[class*="footer"],
.gradio-footer,
.main-footer,
div[class*="footer"],
.show-api,
.built-with-gradio,
a[href*="gradio.app"],
a[href*="huggingface.co/spaces"] {
    display: none !important;
    visibility: hidden !important;
    height: 0 !important;
    padding: 0 !important;
    margin: 0 !important;
}

/* ===== 메인 컨테이너 ===== */
#col-container { 
    max-width: 1200px; 
    margin: 0 auto; 
}

/* ===== 🎨 헤더 타이틀 - 코믹 스타일 ===== */
.header-text h1 {
    font-family: 'Bangers', cursive !important;
    color: #1F2937 !important;
    font-size: 3.5rem !important;
    font-weight: 400 !important;
    text-align: center !important;
    margin-bottom: 0.5rem !important;
    text-shadow: 
        4px 4px 0px #FACC15,
        6px 6px 0px #1F2937 !important;
    letter-spacing: 3px !important;
    -webkit-text-stroke: 2px #1F2937 !important;
}

/* ===== 🎨 서브타이틀 ===== */
.subtitle {
    text-align: center !important;
    font-family: 'Comic Neue', cursive !important;
    font-size: 1.2rem !important;
    color: #1F2937 !important;
    margin-bottom: 1.5rem !important;
    font-weight: 700 !important;
}

/* ===== 🎨 탭 스타일 ===== */
.tabs {
    background: #FFFFFF !important;
    border: 3px solid #1F2937 !important;
    border-radius: 12px !important;
    box-shadow: 6px 6px 0px #1F2937 !important;
    padding: 10px !important;
}

.tab-nav {
    background: #FACC15 !important;
    border-radius: 8px !important;
    padding: 5px !important;
    border: 2px solid #1F2937 !important;
}

.tab-nav button {
    font-family: 'Bangers', cursive !important;
    font-size: 1.1rem !important;
    letter-spacing: 1px !important;
    color: #1F2937 !important;
    background: transparent !important;
    border: none !important;
    padding: 10px 20px !important;
    border-radius: 6px !important;
    transition: all 0.2s ease !important;
}

.tab-nav button:hover {
    background: #FEF3C7 !important;
}

.tab-nav button.selected {
    background: #3B82F6 !important;
    color: #FFFFFF !important;
    box-shadow: 3px 3px 0px #1F2937 !important;
}

/* ===== 🎨 카드/패널 - 만화 프레임 스타일 ===== */
.gr-panel,
.gr-box,
.gr-form,
.block,
.gr-group {
    background: #FFFFFF !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    box-shadow: 6px 6px 0px #1F2937 !important;
    transition: all 0.2s ease !important;
}

.gr-panel:hover,
.block:hover {
    transform: translate(-2px, -2px) !important;
    box-shadow: 8px 8px 0px #1F2937 !important;
}

/* ===== 🎨 입력 필드 (Textbox) ===== */
textarea, 
input[type="text"], 
input[type="number"] {
    background: #FFFFFF !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    color: #1F2937 !important;
    font-family: 'Comic Neue', cursive !important;
    font-size: 1rem !important;
    font-weight: 700 !important;
    transition: all 0.2s ease !important;
}

textarea:focus, 
input[type="text"]:focus, 
input[type="number"]:focus {
    border-color: #3B82F6 !important;
    box-shadow: 4px 4px 0px #3B82F6 !important;
    outline: none !important;
}

textarea::placeholder {
    color: #9CA3AF !important;
    font-weight: 400 !important;
}

/* ===== 🎨 Primary 버튼 - 코믹 블루 ===== */
.gr-button-primary,
button.primary,
.gr-button.primary {
    background: #3B82F6 !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    color: #FFFFFF !important;
    font-family: 'Bangers', cursive !important;
    font-weight: 400 !important;
    font-size: 1.3rem !important;
    letter-spacing: 2px !important;
    padding: 14px 28px !important;
    box-shadow: 5px 5px 0px #1F2937 !important;
    transition: all 0.1s ease !important;
    text-shadow: 1px 1px 0px #1F2937 !important;
}

.gr-button-primary:hover,
button.primary:hover,
.gr-button.primary:hover {
    background: #2563EB !important;
    transform: translate(-2px, -2px) !important;
    box-shadow: 7px 7px 0px #1F2937 !important;
}

.gr-button-primary:active,
button.primary:active,
.gr-button.primary:active {
    transform: translate(3px, 3px) !important;
    box-shadow: 2px 2px 0px #1F2937 !important;
}

/* ===== 🎨 Secondary 버튼 - 코믹 레드 ===== */
.gr-button-secondary,
button.secondary {
    background: #EF4444 !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    color: #FFFFFF !important;
    font-family: 'Bangers', cursive !important;
    font-weight: 400 !important;
    font-size: 1.1rem !important;
    letter-spacing: 1px !important;
    box-shadow: 4px 4px 0px #1F2937 !important;
    transition: all 0.1s ease !important;
    text-shadow: 1px 1px 0px #1F2937 !important;
}

.gr-button-secondary:hover,
button.secondary:hover {
    background: #DC2626 !important;
    transform: translate(-2px, -2px) !important;
    box-shadow: 6px 6px 0px #1F2937 !important;
}

/* ===== 🎨 로그 출력 영역 ===== */
.info-log textarea {
    background: #1F2937 !important;
    color: #10B981 !important;
    font-family: 'Courier New', monospace !important;
    font-size: 0.9rem !important;
    font-weight: 400 !important;
    border: 3px solid #10B981 !important;
    border-radius: 8px !important;
    box-shadow: 4px 4px 0px #10B981 !important;
}

/* ===== 🎨 비디오/오디오 영역 ===== */
.gr-video,
.gr-audio,
video,
audio {
    border: 4px solid #1F2937 !important;
    border-radius: 8px !important;
    box-shadow: 8px 8px 0px #1F2937 !important;
    overflow: hidden !important;
    background: #FFFFFF !important;
}

/* ===== 🎨 갤러리 스타일 ===== */
.gr-gallery {
    background: #FFFFFF !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    box-shadow: 6px 6px 0px #1F2937 !important;
    padding: 10px !important;
}

.gr-gallery .thumbnail-item {
    border: 3px solid #1F2937 !important;
    border-radius: 6px !important;
    transition: all 0.2s ease !important;
    overflow: hidden !important;
}

.gr-gallery .thumbnail-item:hover {
    transform: scale(1.05) !important;
    box-shadow: 4px 4px 0px #3B82F6 !important;
}

/* ===== 🎨 슬라이더 스타일 ===== */
input[type="range"] {
    accent-color: #3B82F6 !important;
}

.gr-slider {
    background: #FFFFFF !important;
}

/* ===== 🎨 라벨 스타일 ===== */
label,
.gr-input-label,
.gr-block-label {
    color: #1F2937 !important;
    font-family: 'Comic Neue', cursive !important;
    font-weight: 700 !important;
    font-size: 1rem !important;
}

span.gr-label {
    color: #1F2937 !important;
}

/* ===== 🎨 정보 텍스트 ===== */
.gr-info,
.info {
    color: #6B7280 !important;
    font-family: 'Comic Neue', cursive !important;
    font-size: 0.9rem !important;
}

/* ===== 🎨 Number Input 스타일 ===== */
.gr-number input {
    background: #FFFFFF !important;
    border: 3px solid #1F2937 !important;
    border-radius: 8px !important;
    color: #1F2937 !important;
    font-family: 'Comic Neue', cursive !important;
    font-weight: 700 !important;
    box-shadow: 3px 3px 0px #1F2937 !important;
}

/* ===== 🎨 스크롤바 - 코믹 스타일 ===== */
::-webkit-scrollbar {
    width: 12px;
    height: 12px;
}

::-webkit-scrollbar-track {
    background: #FEF9C3;
    border: 2px solid #1F2937;
}

::-webkit-scrollbar-thumb {
    background: #3B82F6;
    border: 2px solid #1F2937;
    border-radius: 0px;
}

::-webkit-scrollbar-thumb:hover {
    background: #EF4444;
}

/* ===== 🎨 선택 하이라이트 ===== */
::selection {
    background: #FACC15;
    color: #1F2937;
}

/* ===== 🎨 링크 스타일 ===== */
a {
    color: #3B82F6 !important;
    text-decoration: none !important;
    font-weight: 700 !important;
}

a:hover {
    color: #EF4444 !important;
}

/* ===== 🎨 Row/Column 간격 ===== */
.gr-row {
    gap: 1.5rem !important;
}

.gr-column {
    gap: 1rem !important;
}

/* ===== 반응형 조정 ===== */
@media (max-width: 768px) {
    .header-text h1 {
        font-size: 2.2rem !important;
        text-shadow: 
            3px 3px 0px #FACC15,
            4px 4px 0px #1F2937 !important;
    }
    
    .gr-button-primary,
    button.primary {
        padding: 12px 20px !important;
        font-size: 1.1rem !important;
    }
    
    .gr-panel,
    .block {
        box-shadow: 4px 4px 0px #1F2937 !important;
    }
    
    .tab-nav button {
        font-size: 0.9rem !important;
        padding: 8px 12px !important;
    }
}

/* ===== 🎨 다크모드 비활성화 ===== */
@media (prefers-color-scheme: dark) {
    .gradio-container {
        background-color: #FEF9C3 !important;
    }
}
"""


# Gradio Blocks 인터페이스 생성
with gr.Blocks(fill_height=True, css=css, title="MMAudio Studio") as demo:
    gr.LoginButton(value="Option: HuggingFace 'Login' for extra GPU quota +", size="sm")       
    # HOME Badge
    gr.HTML("""
        <div style="text-align: center; margin: 20px 0 10px 0;">
            <a href="https://www.humangen.ai" target="_blank" style="text-decoration: none;">
                <img src="https://img.shields.io/static/v1?label=🏠 HOME&message=HUMANGEN.AI&color=0000ff&labelColor=ffcc00&style=for-the-badge" alt="HOME">
            </a>
        </div>
    """)
    
    # Header Title
    gr.Markdown(
        """
        # 🎵 MMAUDIO STUDIO 🎬
        """,
        elem_classes="header-text"
    )
    
    gr.Markdown(
        """
        <p class="subtitle">🔊 Generate Audio from Text or Video • Korean Supported! 한글지원 🇰🇷</p>
        """,
    )
    
    with gr.Tabs():
        # Tab 1: Video Search
        with gr.TabItem("🔍 Video Search"):
            gr.Markdown(
                """
                <p style="text-align: center; font-family: 'Comic Neue', cursive; font-weight: 700; color: #1F2937; margin-bottom: 1rem;">
                    📹 Search for videos from Pixabay to use as input!
                </p>
                """
            )
            
            with gr.Row():
                with gr.Column(scale=1):
                    search_query = gr.Textbox(
                        label="🔎 Search Query (한글지원)" if translator else "🔎 Search Query",
                        placeholder="Enter search keywords...",
                        lines=1
                    )
                    search_btn = gr.Button(
                        "🔍 SEARCH VIDEOS!",
                        variant="primary",
                        size="lg"
                    )
                
            search_gallery = gr.Gallery(
                label="📺 Search Results",
                columns=4,
                rows=5,
                height=500
            )
            
            search_btn.click(
                fn=search_videos,
                inputs=[search_query],
                outputs=[search_gallery]
            )
        
        # Tab 2: Video to Audio
        with gr.TabItem("🎬 Video-to-Audio"):
            gr.Markdown(
                """
                <p style="text-align: center; font-family: 'Comic Neue', cursive; font-weight: 700; color: #1F2937; margin-bottom: 1rem;">
                    🎥 Upload a video and generate matching audio!
                </p>
                """
            )
            
            with gr.Row(equal_height=False):
                with gr.Column(scale=1):
                    v2a_video = gr.Video(label="📹 Input Video")
                    v2a_prompt = gr.Textbox(
                        label="✏️ Prompt (한글지원)" if translator else "✏️ Prompt",
                        placeholder="Describe the audio you want...",
                        lines=2
                    )
                    v2a_negative = gr.Textbox(
                        label="🚫 Negative Prompt",
                        value="music",
                        lines=1
                    )
                    
                    with gr.Row():
                        v2a_seed = gr.Number(label="🎲 Seed", value=0)
                        v2a_steps = gr.Number(label="🔄 Steps", value=25)
                    
                    with gr.Row():
                        v2a_cfg = gr.Number(label="🎯 Guidance Scale", value=4.5)
                        v2a_duration = gr.Number(label="⏱️ Duration (sec)", value=8)
                    
                    v2a_btn = gr.Button(
                        "🎬 GENERATE AUDIO! 🔊",
                        variant="primary",
                        size="lg"
                    )
                    
                    with gr.Accordion("📜 Generation Log", open=True):
                        v2a_log = gr.Textbox(
                            label="",
                            placeholder="Upload video and click generate...",
                            lines=12,
                            interactive=False,
                            elem_classes="info-log"
                        )
                
                with gr.Column(scale=1):
                    v2a_output = gr.Video(label="🎥 Generated Result", height=400)
                    gr.Markdown(
                        """
                        <p style="text-align: center; margin-top: 15px; font-weight: 700; color: #1F2937;">
                            💡 Right-click on the video to save!
                        </p>
                        """
                    )
            
            v2a_btn.click(
                fn=video_to_audio,
                inputs=[v2a_video, v2a_prompt, v2a_negative, v2a_seed, v2a_steps, v2a_cfg, v2a_duration],
                outputs=[v2a_output, v2a_log]
            )
        
        # Tab 3: Text to Audio
        with gr.TabItem("🎵 Text-to-Audio"):
            gr.Markdown(
                """
                <p style="text-align: center; font-family: 'Comic Neue', cursive; font-weight: 700; color: #1F2937; margin-bottom: 1rem;">
                    ✨ Generate audio from text description!
                </p>
                """
            )
            
            with gr.Row(equal_height=False):
                with gr.Column(scale=1):
                    t2a_prompt = gr.Textbox(
                        label="✏️ Prompt (한글지원)" if translator else "✏️ Prompt",
                        placeholder="Describe the audio you want to generate...",
                        lines=3
                    )
                    t2a_negative = gr.Textbox(
                        label="🚫 Negative Prompt",
                        placeholder="What to avoid...",
                        lines=1
                    )
                    
                    with gr.Row():
                        t2a_seed = gr.Number(label="🎲 Seed", value=0)
                        t2a_steps = gr.Number(label="🔄 Steps", value=25)
                    
                    with gr.Row():
                        t2a_cfg = gr.Number(label="🎯 Guidance Scale", value=4.5)
                        t2a_duration = gr.Number(label="⏱️ Duration (sec)", value=8)
                    
                    t2a_btn = gr.Button(
                        "🎵 GENERATE AUDIO! ✨",
                        variant="primary",
                        size="lg"
                    )
                    
                    with gr.Accordion("📜 Generation Log", open=True):
                        t2a_log = gr.Textbox(
                            label="",
                            placeholder="Enter prompt and click generate...",
                            lines=12,
                            interactive=False,
                            elem_classes="info-log"
                        )
                
                with gr.Column(scale=1):
                    t2a_output = gr.Audio(label="🔊 Generated Audio")
                    gr.Markdown(
                        """
                        <p style="text-align: center; margin-top: 15px; font-weight: 700; color: #1F2937;">
                            💡 Click the download button to save!
                        </p>
                        """
                    )
            
            t2a_btn.click(
                fn=text_to_audio,
                inputs=[t2a_prompt, t2a_negative, t2a_seed, t2a_steps, t2a_cfg, t2a_duration],
                outputs=[t2a_output, t2a_log]
            )


# 메인 실행
if __name__ == "__main__":
    if translator is None:
        log.warning("Translation model failed to load. Korean translation will be disabled.")
        
    demo.launch(allowed_paths=[output_dir])