app.py

"""
OutofLipSync - Lipsync Only Application
Main Gradio UI module
"""

import os

# Optimize PyTorch memory allocation to reduce fragmentation
os.environ["PYTORCH_ALLOC_CONF"] = "expandable_segments:True"

import logging
import sys
import shutil

import gradio as gr
import torchvision.transforms.functional as _F
from processing import lipsync_with_audio_target
from shared.model_manager import ModelManager


logging.info("=" * 60)
logging.info("APPLICATION STARTING")
logging.info(f"Python version: {sys.version}")
logging.info(f"Platform: {sys.platform}")
logging.info(f"Working directory: {os.getcwd()}")
logging.info("=" * 60)

sys.modules["torchvision.transforms.functional_tensor"] = _F

os.environ["PROCESSED_RESULTS"] = os.path.join(os.getcwd(), "processed_results")
os.makedirs(os.environ["PROCESSED_RESULTS"], exist_ok=True)

src = "/models"
dst = os.path.expanduser("~/.cache/torch/hub/checkpoints")

os.makedirs(dst, exist_ok=True)

if os.path.exists(src):
    for item in os.listdir(src):
        src_path = os.path.join(src, item)
        dst_path = os.path.join(dst, item)
        if os.path.isfile(src_path) and not os.path.exists(dst_path):
            shutil.copy2(src_path, dst_path)
            print(f"Copied {item} to {dst}")

print("Done copying checkpoints!")

print("Loading LatentSync models...")
manager = ModelManager.get_instance()
manager.preload_latentsync_models()
print("Models loaded!")


css = """
    #col-container {
        margin: 0 auto;
        max-width: 1400px;
        padding: 2rem 1rem;
    }
    .header-container {
        background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
        border-radius: 1rem;
        padding: 2rem;
        margin-bottom: 1.5rem;
        box-shadow: 0 4px 20px rgba(102, 126, 234, 0.3);
    }
    .header-title {
        color: white;
        margin: 0;
        font-size: 2.5rem;
        font-weight: 700;
        letter-spacing: -0.5px;
    }
    .header-subtitle {
        color: rgba(255, 255, 255, 0.9);
        margin: 0.5rem 0 0;
        font-size: 1.1rem;
    }
    .card-section {
        background: white;
        border-radius: 1rem;
        padding: 1.5rem;
        box-shadow: 0 2px 12px rgba(0, 0, 0, 0.08);
        border: 1px solid #e5e7eb;
        height: 100%;
        transition: all 0.3s ease;
    }
    .card-section:hover {
        box-shadow: 0 4px 20px rgba(0, 0, 0, 0.12);
    }
    .section-header {
        color: #1f2937;
        font-size: 1.25rem;
        font-weight: 600;
        margin-bottom: 1rem;
        display: flex;
        align-items: center;
        gap: 0.5rem;
    }
    .footer-container {
        margin-top: 2rem;
        padding-top: 1.5rem;
        border-top: 1px solid #e5e7eb;
        text-align: center;
        color: #6b7280;
        font-size: 0.9rem;
    }
    .footer-link {
        color: #667eea;
        text-decoration: none;
        transition: color 0.2s ease;
    }
    .footer-link:hover {
        color: #764ba2;
    }
    """


def cleanup(request: gr.Request):
    sid = request.session_hash
    if sid:
        print(f"{sid} left")
        d1 = os.path.join(os.environ["PROCESSED_RESULTS"], sid)
        shutil.rmtree(d1, ignore_errors=True)


def start_session(request: gr.Request):
    return request.session_hash


with gr.Blocks(css=css) as demo:
    session_state = gr.State()
    demo.load(fn=start_session, outputs=[session_state])

    with gr.Column(elem_id="col-container"):
        gr.HTML(
            """
            <div class="header-container">
                <h1 class="header-title">🎬 OutofLipSync</h1>
                <p class="header-subtitle">Lipsync video with custom audio (English only)</p>
            </div>
            """
        )

        with gr.Row():
            with gr.Column(scale=1):
                with gr.Group(elem_classes="card-section"):
                    gr.HTML('<div class="section-header">📹 Upload Video</div>')
                    video_input = gr.Video(label="Video Source", height=400)

            with gr.Column(scale=1):
                with gr.Group(elem_classes="card-section"):
                    gr.HTML('<div class="section-header">🎵 Upload Audio</div>')
                    audio_input = gr.Audio(
                        label="Target Audio (English only)", type="filepath"
                    )
                    quality_level = gr.Radio(
                        choices=["Fast", "Normal", "Medium", "Best", "Super Best"],
                        value="Normal",
                        label="Quality",
                    )
                    lipsync_only_btn = gr.Button(
                        "👄 Lipsync", variant="primary", size="lg"
                    )

        with gr.Group(elem_classes="card-section"):
            gr.HTML('<div class="section-header">🎬 Final Output</div>')
            final_video = gr.Video(label="Final Output", height=500)

        gr.HTML(
            """
            <div class="footer-container">
                <p>Made with ♥ by <a href="#" class="footer-link">LT Tech</a> • Powered by <a href="#" class="footer-link">LatentSync</a></p>
                <p style="margin-top: 0.5rem; font-size: 0.85rem;">Version 1.0.0</p>
            </div>
            """
        )

    lipsync_only_btn.click(
        fn=lipsync_with_audio_target,
        inputs=[video_input, audio_input, session_state, quality_level],
        outputs=[final_video],
    )


if __name__ == "__main__":
    demo.unload(cleanup)
    demo.queue()
    demo.launch(ssr_mode=False, share=True)