import os
from flask import Flask, request, render_template_string
from PIL import Image
import torch
from transformers import pipeline, CLIPProcessor, CLIPModel
app = Flask(__name__)
# Create the 'static/uploads' folder if it doesn't exist
upload_folder = os.path.join('static', 'uploads')
os.makedirs(upload_folder, exist_ok=True)
# Fake News Detection Models
news_models = {
"mrm8488": pipeline("text-classification", model="mrm8488/bert-tiny-finetuned-fake-news-detection"),
"google-electra": pipeline("text-classification", model="google/electra-base-discriminator"),
"bert-base": pipeline("text-classification", model="bert-base-uncased")
}
# Image Detection Model (CLIP-based)
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
# HTML Template with both Fake News and Image Detection
HTML_TEMPLATE = """
AI & News Detection
📰 Fake News Detection
{% if news_prediction %}
🧠 News Detection Result:
{{ news_prediction }}
{% endif %}
🖼️ AI vs. Human Image Detection
{% if image_prediction %}
📷 Image Detection Result:
{{ image_prediction|safe }}
Explanation: The model compares the uploaded image against the text prompts "AI-generated image" and "Human-created image" to determine similarity. Higher similarity to the AI prompt suggests an AI-generated image, and vice versa.
{% endif %}
"""
@app.route("/", methods=["GET"])
def home():
return render_template_string(HTML_TEMPLATE)
@app.route("/detect", methods=["POST"])
def detect():
text = request.form.get("text")
model_key = request.form.get("model")
if not text or model_key not in news_models:
return render_template_string(HTML_TEMPLATE, news_prediction="Invalid input or model selection.")
result = news_models[model_key](text)[0]
label = "REAL" if result['label'].lower() in ["real", "label_1", "neutral"] else "FAKE"
confidence = result['score'] * 100
prediction_text = f"News is {label} (Confidence: {confidence:.2f}%)"
return render_template_string(HTML_TEMPLATE, news_prediction=prediction_text)
@app.route("/detect_image", methods=["POST"])
def detect_image():
if "image" not in request.files:
return render_template_string(HTML_TEMPLATE, image_prediction="No image uploaded.")
file = request.files["image"]
img = Image.open(file).convert("RGB")
# Compare with AI and Human prompts
prompts = ["AI-generated image", "Human-created image"]
inputs = clip_processor(text=prompts, images=img, return_tensors="pt", padding=True)
with torch.no_grad():
outputs = clip_model(**inputs)
similarity = outputs.logits_per_image.softmax(dim=1).squeeze().tolist()
ai_similarity, human_similarity = similarity
prediction = "AI-Generated" if ai_similarity > human_similarity else "Human-Created"
prediction_text = (
f"Prediction: {prediction} "
f"AI Similarity: {ai_similarity * 100:.2f}% | Human Similarity: {human_similarity * 100:.2f}%"
)
return render_template_string(HTML_TEMPLATE, image_prediction=prediction_text)
if __name__ == "__main__":
app.run(host="0.0.0.0", port=7860)