app.py

import os
import cv2
import glob
import shutil
import numpy as np
import gradio as gr
from PIL import Image
from moviepy import *
from insightface.app import FaceAnalysis

def video_to_images(video_path, images_path):
    cam = cv2.VideoCapture(video_path)
    fps = cam.get(cv2.CAP_PROP_FPS)
    clip = VideoFileClip(video_path)
    clip.audio.write_audiofile("./audio.mp3")

    if os.path.exists(images_path):
        shutil.rmtree(images_path)
    os.makedirs(images_path)

    frame_num = 0
    while True:
        ret, frame = cam.read()
        if not ret:
            break
        filename = os.path.join(images_path, f"{frame_num:05}.png")
        cv2.imwrite(filename, frame, [int(cv2.IMWRITE_PNG_COMPRESSION), 0])
        frame_num += 1

    cam.release()
    return fps

def blur_faces_in_images(images_path):
    faceapp = FaceAnalysis(name="buffalo_l", providers=["CPUExecutionProvider"])
    faceapp.prepare(ctx_id=0)

    for img_path in sorted(glob.glob(os.path.join(images_path, "*.png"))):
        img = cv2.imread(img_path)
        faces = faceapp.get(img)
        for face in faces:
            x1, y1, x2, y2 = list(map(int, face.bbox))
            face_crop = img[y1:y2, x1:x2]
            img[y1:y2, x1:x2] = cv2.GaussianBlur(face_crop, (99, 99), 30)
        cv2.imwrite(img_path, img)

def images_to_video(images_path, output_path, fps):
    images = sorted(glob.glob(os.path.join(images_path, "*.png")))
    frame = cv2.imread(images[0])
    height, width, _ = frame.shape

    temp_path = "temp_video.avi"
    fourcc = cv2.VideoWriter_fourcc(*'XVID')
    out = cv2.VideoWriter(temp_path, fourcc, fps, (width, height))

    for img_path in images:
        frame = cv2.imread(img_path)
        out.write(frame)
    out.release()

    video = VideoFileClip(temp_path)
    audio = AudioFileClip("audio.mp3")
    final = video.set_audio(audio)
    final.write_videofile(output_path, codec="libx264", audio_codec="aac")

    os.remove(temp_path)
    os.remove("audio.mp3")

def process(video):
    input_path = "input.mp4"
    output_path = "output.mp4"
    frames_dir = "frames"

    video.save(input_path)
    fps = video_to_images(input_path, frames_dir)
    blur_faces_in_images(frames_dir)
    images_to_video(frames_dir, output_path, fps)

    shutil.rmtree(frames_dir)
    return output_path

demo = gr.Interface(
    fn=process,
    inputs=gr.Video(label="Upload Video"),
    outputs=gr.Video(label="Faces Blurred"),
    title="Face Blurring App",
    description="Uploads a video and returns the same video with all faces blurred using InsightFace."
)

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