import gradio as gr
import cv2
import numpy as np
import tensorflow as tf

# Load your model
model = tf.keras.models.load_model('model.h5')  # No path needed if it's in the same directory

# Function to resize frames
def resize_frame(frame, size=(64, 64)):
    return cv2.resize(frame, size)

# Function to process each frame
def process_frame(frame):
    # Resize the frame
    resized_frame = resize_frame(frame)

    # Normalize and prepare the frame for the model
    img = resized_frame.astype('float32') / 255.0
    img = np.expand_dims(img, axis=0)
    img = img.reshape(1, -1)  # Flatten to match the input shape

    prediction = model.predict(img)
    return prediction[0][1]  # Assuming category 1 is jumpscare

# Function to convert video to dumpscare
def convert_video_to_dumpscare(video_path, sensitivity):
    cap = cv2.VideoCapture(video_path)
    jumpscare_frames = []

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break
        prediction = process_frame(frame)
        if prediction > sensitivity:  # Adjust this threshold as needed
            jumpscare_frames.append(frame)

    cap.release()

    # Here you can save jumpscare frames or create a new video
    return "Dumpscare video created successfully!"  # Change this as needed

# Gradio interface
def gradio_interface(video, sensitivity):
    result = convert_video_to_dumpscare(video, sensitivity)
    return result

with gr.Blocks() as demo:
    gr.Markdown("## Video Dumpscare Generator")
    video_input = gr.Video(label="Upload Video")
    sensitivity_input = gr.Slider(minimum=0, maximum=1, label="Sensitivity", value=0.5)
    submit_btn = gr.Button("Cut Video")
    output_text = gr.Textbox(label="Output")

    submit_btn.click(gradio_interface, inputs=[video_input, sensitivity_input], outputs=output_text)

demo.launch()