app.py

# app.py

import os
import sys
import time
import gradio as gr
import spaces
from huggingface_hub import snapshot_download
from huggingface_hub.utils import GatedRepoError, RepositoryNotFoundError, RevisionNotFoundError
from pathlib import Path
import tempfile
from pydub import AudioSegment
import cv2
import numpy as np
from scipy import interpolate

# Add the src directory to the system path to allow for local imports
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), 'src')))

from models.inference.moda_test import LiveVASAPipeline, emo_map, set_seed

# --- Configuration ---
# Set seed for reproducibility
set_seed(42)

# Paths and constants for the Gradio demo
DEFAULT_CFG_PATH = "configs/audio2motion/inference/inference.yaml"
DEFAULT_MOTION_MEAN_STD_PATH = "src/datasets/mean.pt"
DEFAULT_SILENT_AUDIO_PATH = "src/examples/silent-audio.wav"
OUTPUT_DIR = "gradio_output"
WEIGHTS_DIR = "pretrain_weights"
REPO_ID = "lixinyizju/moda"

# --- Download Pre-trained Weights from Hugging Face Hub ---
def download_weights():
    """
    Downloads pre-trained weights from Hugging Face Hub if they don't exist locally.
    """
    # A simple check for a key file to see if the download is likely complete
    motion_model_file = os.path.join(WEIGHTS_DIR, "moda", "net-200.pth")
    
    if not os.path.exists(motion_model_file):
        print(f"Weights not found locally. Downloading from Hugging Face Hub repo '{REPO_ID}'...")
        print(f"This may take a while depending on your internet connection.")
        try:
            snapshot_download(
                repo_id=REPO_ID,
                local_dir=WEIGHTS_DIR,
                local_dir_use_symlinks=False,  # Use False to copy files directly; safer for Windows
                resume_download=True,
            )
            print("Weights downloaded successfully.")
        except GatedRepoError:
            raise gr.Error(f"Access to the repository '{REPO_ID}' is gated. Please visit https://huggingface.co/{REPO_ID} to request access.")
        except (RepositoryNotFoundError, RevisionNotFoundError):
            raise gr.Error(f"The repository '{REPO_ID}' was not found. Please check the repository ID.")
        except Exception as e:
            print(f"An error occurred during download: {e}")
            raise gr.Error(f"Failed to download models. Please check your internet connection and try again. Error: {e}")
    else:
        print(f"Found existing weights at '{WEIGHTS_DIR}'. Skipping download.")

# --- Audio Conversion Function ---
def ensure_wav_format(audio_path):
    """
    Ensures the audio file is in WAV format. If not, converts it to WAV.
    Returns the path to the WAV file (either original or converted).
    """
    if audio_path is None:
        return None
    
    audio_path = Path(audio_path)
    
    # Check if already WAV
    if audio_path.suffix.lower() == '.wav':
        print(f"Audio is already in WAV format: {audio_path}")
        return str(audio_path)
    
    # Convert to WAV
    print(f"Converting audio from {audio_path.suffix} to WAV format...")
    
    try:
        # Load the audio file
        audio = AudioSegment.from_file(audio_path)
        
        # Create a temporary WAV file
        with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
            wav_path = tmp_file.name
            # Export as WAV with higher sampling rate for better quality
            audio.export(
                wav_path,
                format='wav',
                parameters=["-ar", "24000", "-ac", "1"]  # 24kHz, mono for better lip-sync
            )
        
        print(f"Audio converted successfully to: {wav_path}")
        return wav_path
        
    except Exception as e:
        print(f"Error converting audio: {e}")
        raise gr.Error(f"Failed to convert audio file to WAV format. Error: {e}")

# --- Frame Interpolation Function ---
def interpolate_frames(video_path, target_fps=30):
    """
    Interpolates frames in a video to achieve smoother motion.
    
    Args:
        video_path: Path to the input video
        target_fps: Target frames per second
    
    Returns:
        Path to the interpolated video
    """
    try:
        video_path = str(video_path)
        cap = cv2.VideoCapture(video_path)
        
        # Get original video properties
        original_fps = cap.get(cv2.CAP_PROP_FPS)
        width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        
        # Fix for FPS detection issue
        if original_fps == 0 or original_fps is None:
            print("Warning: Could not detect original FPS. Assuming 25 FPS.")
            original_fps = 25.0
        
        print(f"Original FPS: {original_fps}, Target FPS: {target_fps}")
        
        # If target FPS is not higher, return original
        if original_fps >= target_fps:
            cap.release()
            print("Target FPS is not higher than original. Skipping interpolation.")
            return video_path
        
        # Read all frames
        frames = []
        while True:
            ret, frame = cap.read()
            if not ret:
                break
            frames.append(frame)
        cap.release()
        
        if len(frames) < 2:
            print("Not enough frames for interpolation.")
            return video_path
        
        # Calculate interpolation factor (can be fractional)
        interpolation_factor = target_fps / original_fps
        
        # For fractional factors, we need different approach
        if interpolation_factor <= 1:
            print("Interpolation factor too low. Skipping.")
            return video_path
            
        print(f"Interpolating with factor: {interpolation_factor:.2f}")
        print(f"Total frames to process: {len(frames)}")
        
        # Perform frame interpolation
        interpolated_frames = []
        
        if interpolation_factor == int(interpolation_factor):
            # Integer factor - simple interpolation
            factor = int(interpolation_factor)
            for i in range(len(frames) - 1):
                interpolated_frames.append(frames[i])
                # Generate intermediate frames
                for j in range(1, factor):
                    alpha = j / factor
                    interpolated_frame = cv2.addWeighted(
                        frames[i], 1 - alpha,
                        frames[i + 1], alpha,
                        0
                    )
                    interpolated_frames.append(interpolated_frame)
            interpolated_frames.append(frames[-1])
        else:
            # Fractional factor - use different approach
            # For 25 -> 60 fps, we need to add selective frames
            for i in range(len(frames) - 1):
                interpolated_frames.append(frames[i])
                # Add intermediate frame for smoother motion
                if i % 2 == 0:  # Add extra frame every other original frame
                    alpha = 0.4  # Blend ratio
                    interpolated_frame = cv2.addWeighted(
                        frames[i], 1 - alpha,
                        frames[i + 1], alpha,
                        0
                    )
                    interpolated_frames.append(interpolated_frame)
            interpolated_frames.append(frames[-1])
        
        print(f"Total interpolated frames: {len(interpolated_frames)}")
        
        # Save the interpolated video
        output_path = video_path.replace('.mp4', '_interpolated.mp4')
        
        # Use H.264 codec for better compatibility
        fourcc = cv2.VideoWriter_fourcc(*'H264')
        out = cv2.VideoWriter(output_path, fourcc, target_fps, (width, height))
        
        if not out.isOpened():
            # Fallback to mp4v if H264 not available
            fourcc = cv2.VideoWriter_fourcc(*'mp4v')
            out = cv2.VideoWriter(output_path, fourcc, target_fps, (width, height))
        
        for frame in interpolated_frames:
            out.write(frame)
        out.release()
        
        print(f"Interpolated video saved to: {output_path}")
        return output_path
        
    except Exception as e:
        print(f"Error during frame interpolation: {e}")
        import traceback
        traceback.print_exc()
        return video_path  # Return original if interpolation fails

# --- Initialization ---
# Create output directory if it doesn't exist
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Download weights before initializing the pipeline
download_weights()

# Instantiate the pipeline once to avoid reloading models on every request
print("Initializing MoDA pipeline...")
try:
    pipeline = LiveVASAPipeline(
        cfg_path=DEFAULT_CFG_PATH,
        motion_mean_std_path=DEFAULT_MOTION_MEAN_STD_PATH
    )
    print("MoDA pipeline initialized successfully.")
except Exception as e:
    print(f"Error initializing pipeline: {e}")
    pipeline = None

# Invert the emo_map for easy lookup from the dropdown value
emo_name_to_id = {v: k for k, v in emo_map.items()}

# --- Audio Length Check Function ---
def check_audio_length(audio_path):
    """
    Check the length of an audio file and warn if it's too long.
    
    Args:
        audio_path: Path to the audio file
    
    Returns:
        Duration in seconds
    """
    try:
        audio = AudioSegment.from_file(audio_path)
        duration_seconds = len(audio) / 1000.0
        return duration_seconds
    except Exception as e:
        print(f"Error checking audio length: {e}")
        return None

# --- Core Generation Function ---
@spaces.GPU(duration=180)  # Increased duration for smoothing and interpolation
def generate_motion(source_image_path, driving_audio_path, emotion_name, 
                   cfg_scale, smooth_enabled, target_fps,
                   progress=gr.Progress(track_tqdm=True)):
    """
    The main function that takes Gradio inputs and generates the talking head video.
    
    Args:
        source_image_path: Path to the source image
        driving_audio_path: Path to the driving audio
        emotion_name: Selected emotion
        cfg_scale: CFG scale for generation
        smooth_enabled: Whether to enable smoothing
        target_fps: Target frames per second for interpolation
    """
    if pipeline is None:
        raise gr.Error("Pipeline failed to initialize. Check the console logs for details.")
        
    if source_image_path is None:
        raise gr.Error("Please upload a source image.")
    if driving_audio_path is None:
        raise gr.Error("Please upload a driving audio file.")
    
    # Check audio length
    audio_duration = check_audio_length(driving_audio_path)
    if audio_duration:
        print(f"Audio duration: {audio_duration:.1f} seconds")
        if audio_duration > 60:
            gr.Warning(f"⚠️ Audio is {audio_duration:.1f} seconds long. MoDA works best with audio under 60 seconds. Processing may be slow and quality may degrade.")
        if audio_duration > 180:
            raise gr.Error("Audio is too long. Please use audio files under 3 minutes (180 seconds) for best results.")

    start_time = time.time()
    
    # Ensure audio is in WAV format with optimal sampling rate
    wav_audio_path = ensure_wav_format(driving_audio_path)
    temp_wav_created = wav_audio_path != driving_audio_path
    
    # Create a unique subdirectory for this run
    timestamp = time.strftime("%Y%m%d-%H%M%S")
    run_output_dir = os.path.join(OUTPUT_DIR, timestamp)
    os.makedirs(run_output_dir, exist_ok=True)

    # Get emotion ID from its name
    emotion_id = emo_name_to_id.get(emotion_name, 8)  # Default to 'None' (ID 8) if not found

    print(f"Starting generation with the following parameters:")
    print(f"  Source Image: {source_image_path}")
    print(f"  Driving Audio (original): {driving_audio_path}")
    print(f"  Driving Audio (WAV): {wav_audio_path}")
    print(f"  Emotion: {emotion_name} (ID: {emotion_id})")
    print(f"  CFG Scale: {cfg_scale}")
    print(f"  Smoothing: {smooth_enabled}")
    print(f"  Target FPS: {target_fps}")

    try:
        # Temporarily disable smoothing if it causes CUDA tensor issues
        # Check if smooth causes issues and handle gracefully
        try:
            # Try with smoothing first
            result_video_path = pipeline.driven_sample(
                image_path=source_image_path,
                audio_path=wav_audio_path,
                cfg_scale=float(cfg_scale),
                emo=emotion_id,
                save_dir=".",
                smooth=smooth_enabled,  # Use the checkbox value
                silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
            )
        except TypeError as tensor_error:
            if "can't convert cuda" in str(tensor_error) and smooth_enabled:
                print("Warning: Smoothing caused CUDA tensor error. Retrying without smoothing...")
                # Retry without smoothing
                result_video_path = pipeline.driven_sample(
                    image_path=source_image_path,
                    audio_path=wav_audio_path,
                    cfg_scale=float(cfg_scale),
                    emo=emotion_id,
                    save_dir=".",
                    smooth=False,  # Disable smoothing as fallback
                    silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
                )
                print("Generated video without smoothing due to technical limitations.")
            else:
                raise tensor_error
        
        # Apply frame interpolation if requested
        if target_fps > 24:  # Assuming default is around 24 FPS
            print(f"Applying frame interpolation to achieve {target_fps} FPS...")
            result_video_path = interpolate_frames(result_video_path, target_fps=target_fps)
        
    except Exception as e:
        print(f"An error occurred during video generation: {e}")
        import traceback
        traceback.print_exc()
        raise gr.Error(f"An unexpected error occurred: {str(e)}. Please check the console for details.")
    finally:
        # Clean up temporary WAV file if created
        if temp_wav_created and os.path.exists(wav_audio_path):
            try:
                os.remove(wav_audio_path)
                print(f"Cleaned up temporary WAV file: {wav_audio_path}")
            except Exception as e:
                print(f"Warning: Could not delete temporary file {wav_audio_path}: {e}")

    end_time = time.time()
    processing_time = end_time - start_time
    
    result_video_path = Path(result_video_path)
    final_path = result_video_path.with_name(f"final_{result_video_path.stem}{result_video_path.suffix}")
    
    print(f"Video generated successfully at: {final_path}")
    print(f"Processing time: {processing_time:.2f} seconds.")

    return final_path

# --- Gradio UI Definition ---
with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 960px !important; margin: 0 auto !important}") as demo:
    gr.HTML(
        """
        <div align='center'>
            <h1>MoDA: Multi-modal Diffusion Architecture for Talking Head Generation</h1>
            <h2 style="color: #4A90E2;">Enhanced Version with Smooth Motion</h2>
            <p style="display:flex; justify-content: center; gap: 10px;">
                <a href='https://lixinyyang.github.io/MoDA.github.io/'><img src='https://img.shields.io/badge/Project-Page-blue'></a>
                <a href='https://arxiv.org/abs/2507.03256'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a>
                <a href='https://github.com/lixinyyang/MoDA/'><img src='https://img.shields.io/badge/Code-Github-green'></a>
            </p>
        </div>
        """
    )

    with gr.Row(variant="panel"):
        with gr.Column(scale=1):
            gr.Markdown("### 📥 Input Settings")
            
            with gr.Row():
                source_image = gr.Image(
                    label="Source Image", 
                    type="filepath", 
                    value="src/examples/reference_images/7.jpg"
                )
            
            with gr.Row():
                driving_audio = gr.Audio(
                    label="Driving Audio (Recommended: < 60 seconds)",
                    type="filepath",
                    value="src/examples/driving_audios/5.wav"
                )

            gr.Markdown("### ⚙️ Generation Settings")
            
            with gr.Row():
                emotion_dropdown = gr.Dropdown(
                    label="Emotion",
                    choices=list(emo_map.values()),
                    value="None",
                    info="Select an emotion for more natural facial expressions"
                )

            with gr.Row():
                cfg_slider = gr.Slider(
                    label="CFG Scale (Lower = Smoother motion)",
                    minimum=0.5,
                    maximum=5.0,
                    step=0.1,
                    value=0.5,
                    info="Lower values produce smoother but less controlled motion"
                )
            
            gr.Markdown("### 🎬 Motion Enhancement")
            
            with gr.Row():
                smooth_checkbox = gr.Checkbox(
                    label="Enable Smoothing (Experimental)",
                    value=False,  # Changed to False due to CUDA issues
                    info="May cause errors on some systems. If errors occur, disable this option."
                )
            
            with gr.Row():
                fps_slider = gr.Slider(
                    label="Target FPS",
                    minimum=24,
                    maximum=50,
                    step=1,
                    value=50,
                    info="Higher FPS for smoother motion. 30 FPS recommended, 50 FPS maximum"
                )
            
            submit_button = gr.Button("🎥 Generate Video", variant="primary", size="lg")

        with gr.Column(scale=1):
            gr.Markdown("### 📺 Output")
            output_video = gr.Video(label="Generated Video")
            
            # Processing status
            with gr.Row():
                gr.Markdown(
                    """
                    <div style="background-color: #f0f8ff; padding: 10px; border-radius: 5px; margin-top: 10px;">
                        <p style="margin: 0; font-size: 0.9em;">
                        <b>Tips for best results:</b><br>
                        • Use high-quality front-facing images<br>
                        • Clear audio without background noise<br>
                        • <b>Keep audio under 60 seconds</b><br>
                        • Adjust CFG scale if motion seems stiff<br>
                        • For longer audio, split into segments
                        </p>
                    </div>
                    """
                )

    gr.Markdown(
        """
        ---
        ### ⚠️ **Disclaimer**
        This project is intended for academic research, and we explicitly disclaim any responsibility for user-generated content. 
        Users are solely liable for their actions while using this generative model.
        
        ### 🚀 **Enhancement Features**
        - **Frame Smoothing**: Reduces jitter and improves transition between frames (currently experimental)
        - **Frame Interpolation**: Increases FPS for smoother motion
        - **Optimized Audio Processing**: Better lip-sync with 24kHz sampling
        - **Fine-tuned CFG Scale**: Better control over motion naturalness
        
        ### ⏱️ **Audio Length Limitations**
        - **Optimal**: Under 30 seconds for best quality and speed
        - **Recommended**: Under 60 seconds 
        - **Maximum**: 180 seconds (3 minutes) - very slow processing
        - For longer content, consider splitting audio into segments
        """
    )
    
    # Examples section
    gr.Examples(
        examples=[
            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "None", 1.0, False, 30],
            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Happy", 0.8, False, 30],
            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Sad", 1.2, False, 24],
        ],
        inputs=[source_image, driving_audio, emotion_dropdown, cfg_slider, smooth_checkbox, fps_slider],
        outputs=output_video,
        fn=generate_motion,
        cache_examples=False,
        label="Example Configurations"
    )
    
    submit_button.click(
        fn=generate_motion,
        inputs=[source_image, driving_audio, emotion_dropdown, cfg_slider, smooth_checkbox, fps_slider],
        outputs=output_video
    )

if __name__ == "__main__":
    demo.launch(share=True)