Update app.py

app.py

@@ -2,6 +2,7 @@ import os
 import cv2
 import numpy as np
 import gradio as gr
+from PIL import Image
 import tensorflow as tf
 from tensorflow.keras.models import load_model
 from tensorflow.keras.applications.xception import preprocess_input as xcp_pre
@@ -14,31 +15,33 @@ eff_path = hf_hub_download(repo_id="Zeyadd-Mostaffa/deepfake-image-detector", fi
 xcp_model = load_model(xcp_path)
 eff_model = load_model(eff_path)
 
-def predict(img):
+def predict(image_pil: Image.Image) -> str:
     try:
-        # Resize input
-        xcp_img = cv2.resize(img, (299, 299))
-        eff_img = cv2.resize(img, (224, 224))
+        # Convert PIL to numpy
+        image = np.array(image_pil.convert("RGB"))
+
+        # Resize and preprocess
+        xcp_img = cv2.resize(image, (299, 299))
+        eff_img = cv2.resize(image, (224, 224))
 
-        # Preprocess for each model
         xcp_tensor = xcp_pre(xcp_img.astype(np.float32))[np.newaxis, ...]
         eff_tensor = eff_pre(eff_img.astype(np.float32))[np.newaxis, ...]
 
         # Predict
         xcp_pred = xcp_model.predict(xcp_tensor, verbose=0).flatten()[0]
         eff_pred = eff_model.predict(eff_tensor, verbose=0).flatten()[0]
-
         avg_pred = (xcp_pred + eff_pred) / 2
-        return "Real" if avg_pred > 0.5 else "Fake"
 
+        return "Real" if avg_pred > 0.5 else "Fake"
     except Exception as e:
-        return "Error: " + str(e)
+        return f"Error: {str(e)}"
 
-# ✅ Use literal type-safe components
 demo = gr.Interface(
     fn=predict,
-    inputs=gr.Image(type="numpy"),
-    outputs=gr.Textbox(),
+    inputs=gr.Image(type="pil", label="Upload Image"),  # ✅ Use PIL instead of numpy
+    outputs=gr.Textbox(label="Prediction"),             # ✅ Safe, schema-compatible
+    title="Deepfake Image Detector",
+    description="Upload a full image. The model classifies it as real or fake.",
     allow_flagging="never"
 )