app.py

import gradio as gr
import torch
from PIL import Image as PILImage
from transformers import AutoImageProcessor, SiglipForImageClassification

# --- Model Loading ---
MODEL_IDENTIFIER = r"Ateeqq/ai-vs-human-image-detector"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

print(f"Loading model: {MODEL_IDENTIFIER} on device: {device}")
processor = AutoImageProcessor.from_pretrained(MODEL_IDENTIFIER)
model = SiglipForImageClassification.from_pretrained(MODEL_IDENTIFIER)
model.to(device)
model.eval()
print("Model loaded successfully.")

# --- Prediction Function ---
def predict(image):
    """
    Takes a PIL image, preprocesses it, and returns the prediction probabilities.
    """
    if image is None:
        return None
        
    # Preprocess the image
    inputs = processor(images=image, return_tensors="pt").to(device)

    # Perform inference
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        
    # Get probabilities
    probabilities = torch.softmax(logits, dim=-1)[0]
    
    # Create a dictionary of labels and their scores
    confidences = {model.config.id2label[i]: score.item() for i, score in enumerate(probabilities)}
    return confidences

# --- Gradio Interface ---
# Define the Gradio interface components
image_input = gr.Image(type="pil", label="Upload an Image")
label_output = gr.Label(num_top_classes=2, label="Prediction")

# The title and description for the app
title = "AI vs Human Image Detector"
description = """
This Space uses the `Ateeqq/ai-vs-human-image-detector` model to classify an image as either AI-generated or Human-made. 
Upload an image to see the prediction.
"""
article = "Model by [Ateeqq](https://huggingface.co/Ateeqq) | Gradio app created with AI"

# Launch the Gradio app
gr.Interface(
    fn=predict,
    inputs=image_input,
    outputs=label_output,
    title=title,
    description=description,
    article=article
).launch(share=True, server_name="0.0.0.0")