Powered by EfficientNet-B0

Is That Image Real or AI-Generated?

Upload any image and our deep learning model will analyze it in seconds. Built with transfer learning on EfficientNet-B0 for high-accuracy detection.

Upload & Detect
94.7% Accuracy
<2s Inference
100% Local & Private

How It Works

Three simple steps to verify any image

1. Upload

Drag & drop or click to upload a JPG, PNG, or WebP image from your device.

2. Analyze

Our EfficientNet-B0 model processes the image through deep neural network layers to detect AI artifacts.

3. Result

Get an instant verdict with a confidence score — see exactly how sure the model is about its prediction.

Try the Detector

Your image never leaves your computer — all processing happens locally

Drag & drop your image here or click to browse from your device Supports JPG, PNG, WebP

Under the Hood

Built with industry-standard deep learning tools

🧠

EfficientNet-B0

Pre-trained on ImageNet with fine-tuned layers for AI image detection

🔥

PyTorch

Training pipeline with transfer learning, cosine LR scheduling & early stopping

FastAPI

High-performance async Python backend serving predictions in milliseconds

🔒

100% Local

No cloud uploads — your images are processed entirely on your machine

Frequently Asked Questions

What types of AI-generated images can it detect?

The model is trained to detect images created by diffusion models (Stable Diffusion, DALL-E, Midjourney) and GANs (StyleGAN, etc.). Detection accuracy depends on the diversity and quality of the training dataset.

How accurate is the detection?

The current model achieves ~94.7% validation accuracy. Accuracy improves with more training data. For best results, use diverse images of similar resolution to the training set.

Is my image uploaded to the cloud?

No. Everything runs locally on your machine. The image is sent to a local server (localhost) and is never transmitted over the internet.

What image formats are supported?

JPG/JPEG, PNG, and WebP. The image is automatically resized and normalized before inference — any resolution works.

Can I improve the model's accuracy?

Yes! Add more images to the data/train/ and data/val/ folders, then retrain with python -m training.train. More diverse data leads to better generalization.