colab_flask.py

import os
import subprocess
import time
from datetime import datetime

import cv2
from flask import Flask, request, jsonify, send_file
from flask_ngrok2 import run_with_ngrok
from subprocess import call
from tqdm import tqdm
import numpy as np
from ffmpy import FFmpeg

# !pip install flask flask-ngrok2 pyngrok

app = Flask(__name__)
auth_token = '2XQTtjJqVg4B4ryQVgauTPIGeiK_3JUWVJcMhXwpPxz2sc9KB'
root_dir = '/content/wav2lip-gfpgan'
jobs_dir = os.path.join('/content', 'jobs')


@app.route('/', methods=['GET'])
def index():
    return jsonify({'hello': 'world!'})


# New route to serve files from /content/input
@app.route('/job/<job_id>/<filename>', methods=['GET'])
def download_file(job_id, filename):
    try:
        file_path = os.path.join(jobs_dir, job_id, filename)
        print(file_path)
        return send_file(file_path, as_attachment=True)
    except Exception as e:
        return jsonify({'error': str(e)}), 500


@app.route('/wav2lip', methods=['POST'])
def wav2lip():
    try:
        # Get uploaded files
        video_file = request.files['video']
        audio_file = request.files['audio']
        job_id = str(int(time.time()))
        job_path = os.path.join(jobs_dir, job_id)
        os.makedirs(job_path, exist_ok=True)

        # Save the files to a temporary directory
        video_path = os.path.join(job_path, video_file.filename)
        audio_path = os.path.join(job_path, audio_file.filename)
        video_file.save(video_path)
        audio_file.save(audio_path)

        wav2lip_mp4 = os.path.join(job_path, 'wav2lip.mp4')
        call_wav2lip(video_path, audio_path, wav2lip_mp4)

        output_filename = 'output.mp4'
        output_mp4 = os.path.join(job_path, output_filename)
        call_gfpgan(wav2lip_mp4, audio_path, output_mp4)

        return jsonify({'url': f'/job/{job_id}/{output_filename}'})

    except Exception as e:
        return jsonify({'error': str(e)}), 500


def call_wav2lip(video_path, audio_path, output_path):
    checkpoint_path = os.path.join(root_dir, 'wav2lip/checkpoints/wav2lip.pth')
    assert os.path.isfile(video_path), f'Video path {video_path} not exist.'
    assert os.path.isfile(audio_path), f'Audio path {audio_path} not exist.'
    assert os.path.isfile(checkpoint_path), f'Checkpoint file {checkpoint_path} not exist.'

    # python inference.py \
    # --checkpoint_path checkpoints/wav2lip.pth \
    # --face {inputVideoPath} \
    # --audio {inputAudioPath} \
    # --outfile {lipSyncedOutputPath}
    start = datetime.now()
    cmd = [
        "python",
        "wav2lip/inference.py",
        "--checkpoint_path", checkpoint_path,  #
        # "--segmentation_path", "checkpoints/face_segmentation.pth",
        "--face", video_path,
        "--audio", audio_path,
        "--outfile", output_path,
    ]
    print(f'Call subprocess: {cmd}')
    proc = subprocess.Popen(cmd, shell=False)
    proc.communicate()
    duration = datetime.now() - start
    print(f'wav2lip finished in {duration}')
    return output_path


def _get_frames(video_path):
    folder_path = os.path.dirname(video_path)
    origin_frames_folder = os.path.join(folder_path, 'frames')
    os.makedirs(origin_frames_folder, exist_ok=True)

    # get frames pics
    vidcap = cv2.VideoCapture(video_path)
    numberOfFrames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
    fps = vidcap.get(cv2.CAP_PROP_FPS)
    print("FPS: ", fps, "Frames: ", numberOfFrames)

    for frameNumber in tqdm(range(numberOfFrames)):
        _, image = vidcap.read()
        cv2.imwrite(os.path.join(origin_frames_folder, str(frameNumber).zfill(4) + '.jpg'), image)
    return origin_frames_folder


def call_gfpgan(wav2lip_mp4, audio_path, output_mp4):
    assert os.path.isfile(wav2lip_mp4), f'Video path {wav2lip_mp4} not exist.'
    origin_frames_folder = _get_frames(wav2lip_mp4)
    folder_path = os.path.dirname(wav2lip_mp4)

    # python inference_gfpgan.py
    # -i "$unProcessedFramesFolderPath"
    # -o "$outputPath"
    # -v 1.3
    # -s 2
    # --only_center_face
    # --bg_upsampler None
    start = datetime.now()
    cmd = [
        "python",
        "gfpgan/inference_gfpgan.py",
        "-i", origin_frames_folder,
        "-o", folder_path,
        # "-v", str(1.4),
        # "-s", str(2),
        "--only_center_face",
        "--bg_upsampler", 'realesrgan'
    ]
    print(cmd)
    proc = subprocess.Popen(cmd, shell=True)
    proc.communicate()
    duration = datetime.now() - start
    print(f'inference_gfpgan finished in {duration}')

    start = datetime.now()
    cmd = [
        "python",
        "merge.py",
        "-j", folder_path,
        "-a", audio_path,
        "-o", output_mp4,
    ]
    proc = subprocess.Popen(cmd, shell=True)
    proc.communicate()
    duration = datetime.now() - start
    print(f'Merge output in {duration}')
    print(output_mp4)

    # from google.colab import files
    # files.download(finalProcessedOuputVideo)


if __name__ == '__main__':
    run_with_ngrok(app, auth_token=auth_token)
    app.run()