app.py

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

def preprocess_image(image):
    image = cv2.resize(image, (224, 224))  # Resize to model input size
    image = image / 255.0  # Normalize to [0,1] range
    image = np.expand_dims(image, axis=0)  # Add batch dimension
    return image

# Load trained model
model = tf.keras.models.load_model("xception_deepfake_image.h5")

def predict_deepfake(image):
    image = preprocess_image(image)
    prediction = model.predict(image)[0][0]  # Model outputs probability
    label = "FAKE" if prediction > 0.5 else "REAL"
    confidence = prediction if label == "FAKE" else 1 - prediction
    return {"REAL": float(1 - prediction), "FAKE": float(prediction)}

# Create Gradio Interface
demo = gr.Interface(
    fn=predict_deepfake,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(num_top_classes=2),
    title="DeepFake Image Detector",
    description="Upload an image to check if it's REAL or FAKE",
)

demo.launch()