// Analyze
Run a Scan
Upload an image or video to check for signs of AI-generated or manipulated content.
⬆
Drop file here or click to upload
Images · Videos · Max 100MB
JPGPNGWEBPMP4MOVAVIMKV
Advanced deepfake detection for images and videos — powered by a fine-tuned EfficientNet-B4 classifier with ~99% validation accuracy. Upload media and get an instant authenticity verdict with confidence scoring.
Upload an image or video to check for signs of AI-generated or manipulated content.
Live analytics from your scan history this session.
A deepfake detection system built end-to-end, from dataset to deployment.
Forensa uses a fine-tuned EfficientNet-B4 backbone with a custom classification head, trained on 140K real and fake face images. The model achieved ~99% validation accuracy on a held-out test set after training on a Tesla T4 GPU.
For images, the model analyzes pixel-level artifacts and inconsistencies typical of GAN-generated content. For videos, frames are sampled at a configurable rate, each frame is scored independently, and results are averaged to produce a final verdict with a confidence score.
Trained on a curated dataset of 140,000 real and fake face images sourced from Kaggle, with careful class-balance handling and preprocessing to ensure robust generalization across different face types, lighting conditions, and compression artifacts.
Forensa was built as an independent research project covering the full ML lifecycle — dataset preprocessing, model training, evaluation, and deployment as a live web app.