import gradio as gr
import requests
from PIL import Image
from transformers import pipeline, Pipeline
import os
from dotenv import load_dotenv

# --- Configuration ---
# Load secrets from the Space's "Repository secrets" settings
load_dotenv()
VALID_BEARER_TOKEN = os.getenv("VALID_BEARER_TOKEN")
OWNER_PHONE_NUMBER = os.getenv("OWNER_PHONE_NUMBER")

# --- AI Model Setup ---
# This is loaded once when the Space starts
print("Loading AI Image Detection model...")
image_detector: Pipeline = pipeline("image-classification", model="openai/clip-vit-base-patch32")
print("✅ Model loaded successfully.")

# --- Main Tool Function ---
def analyze_image_authenticity(image_url: str) -> dict:
    """
    Analyzes an image from a URL to determine if it is real or AI-generated.
    
    Args:
        image_url: The URL of the image to analyze.
        
    Returns:
        A dictionary with the analysis results and probability scores.
    """
    if not image_url:
        raise gr.Error("Image URL parameter is missing.")
    
    print(f"Analyzing image from URL: {image_url}")
    try:
        image = Image.open(requests.get(image_url, stream=True, timeout=10).raw)
    except Exception as e:
        # For Gradio, it's better to raise a gr.Error for user-facing issues
        raise gr.Error(f"Could not load image from URL. It might be invalid or inaccessible. Error: {str(e)}")

    labels = ["a real photograph", "a computer-generated image", "an illustration or drawing"]
    results = image_detector(image, candidate_labels=labels)
    
    print(f"Analysis successful. Results: {results}")
    return {"analysis_results": results}

# --- Gradio Interface ---
# This defines the UI and the MCP endpoint
demo = gr.Interface(
    fn=analyze_image_authenticity,
    inputs=[gr.Textbox(label="Image URL")],
    outputs=[gr.JSON(label="Analysis Results")],
    title="AI Image Authenticity Detector",
    description="Provide an image URL to determine if it is a real photograph or AI-generated."
)

# --- Launch the App and MCP Server ---
# mcp_server=True is the magic parameter that exposes your function as an MCP tool
demo.launch(mcp_server=True)