2.5M real and synthetic images across 25 generation methods, including GANs and diffusion models — used to train and evaluate Integrity's quick and deep scan models.

Enhances deepfake detection generalizability using a ResNet50 model with Gaussian blur preprocessing across multiple dataset splits.

Compares eight CNN and Transformer architectures for deepfake detection across multiple second- and third-generation deepfake datasets.

A Streamlit app combining a ResNet CNN with an LSTM to detect deepfake videos by learning spatial and temporal patterns across frames.

Deep learning framework used for deep scan model training and inference.

Lightweight inference engine powering the quick scan tier's instant predictions.

Open standard for content provenance and authenticity metadata, scanned by the quick scan's metadata branch.

Full source code, model weights, training scripts, and API documentation.