import gradio as gr
from transformers import pipeline, CLIPProcessor, CLIPModel
from PIL import Image
import torch
import cv2
import os
import tempfile

# Load models
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
news_model = pipeline("text-classification", model="mrm8488/bert-tiny-finetuned-fake-news")
# AI Image Detection
def analyze_image(image):
    inputs = clip_processor(text=["a real photo", "an AI-generated image"], images=image, return_tensors="pt", padding=True)
    outputs = clip_model(**inputs)
    logits_per_image = outputs.logits_per_image
    probs = logits_per_image.softmax(dim=1).tolist()[0]
    prediction = "AI-generated" if probs[1] > probs[0] else "Real"
    confidence = round(max(probs) * 100, 2)

    explanation = (
        "The image strongly suggests AI generation. Look for smooth textures, perfect symmetry, "
        "and overly clean details — signs of AI creation."
        if prediction == "AI-generated"
        else "This image appears authentic based on texture, lighting, and organic imperfections. "
             "No major AI traits were detected."
    )
    return f"Prediction: {prediction} ({confidence}%)", explanation

# Fake News Detection
def analyze_news(text):
    result = news_model(text)[0]
    label = result["label"]
    score = round(result["score"] * 100, 2)

    pred = "Fake News" if label == "LABEL_0" else "Real News"
    explanation = (
        "⚠️ This article likely contains misinformation or fabricated facts."
        if label == "LABEL_0"
        else "✅ The article seems factual and based on credible information."
    )

    return f"Prediction: {pred} ({score}%)", explanation

# Video Analysis
def analyze_video(video_file):
    temp_dir = tempfile.mkdtemp()
    video_path = os.path.join(temp_dir, "input.mp4")
    with open(video_path, "wb") as f:
        f.write(video_file.read())

    cap = cv2.VideoCapture(video_path)
    fps = cap.get(cv2.CAP_PROP_FPS)
    frame_rate = int(fps) if fps > 0 else 1

    results = []
    frame_count = 0

    while cap.isOpened():
        success, frame = cap.read()
        if not success:
            break

        if frame_count % frame_rate == 0:
            pil_image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
            result, _ = analyze_image(pil_image)
            results.append(result)

        frame_count += 1

    cap.release()

    if any("AI-generated" in r for r in results):
        return "Prediction: AI-generated content detected in video.", (
            "The video includes frames that strongly resemble AI-generated imagery. "
            "Check for smooth rendering, cartoonish features, or lack of natural flaws."
        )
    else:
        return "Prediction: Video looks real.", (
            "No strong signs of AI generation were detected in any frame. "
            "Visuals appear natural and consistent with real-world content."
        )

# Gradio App
with gr.Blocks() as demo:
    gr.Markdown("## 🧠 Fake News & AI Media Detector (Image / News / Video)")

    with gr.Row():
        with gr.Column():
            img_input = gr.Image(label="Upload Image")
            img_btn = gr.Button("Analyze Image")
            img_result = gr.Textbox(label="Image Result")
            img_explanation = gr.Textbox(label="Image Explanation")

        with gr.Column():
            news_input = gr.Textbox(label="Paste News Headline or Text")
            news_btn = gr.Button("Analyze News")
            news_result = gr.Textbox(label="News Result")
            news_explanation = gr.Textbox(label="News Explanation")

    with gr.Row():
        video_input = gr.File(label="Upload Short Video (≤ 10 sec, MP4)", file_types=[".mp4"])
        video_btn = gr.Button("Analyze Video")
        video_result = gr.Textbox(label="Video Result")
        video_explanation = gr.Textbox(label="Video Explanation")

    img_btn.click(analyze_image, inputs=img_input, outputs=[img_result, img_explanation])
    news_btn.click(analyze_news, inputs=news_input, outputs=[news_result, news_explanation])
    video_btn.click(analyze_video, inputs=video_input, outputs=[video_result, video_explanation])

demo.launch()