FAKES FOR ALL: Demographically Inclusive Face-Swapped Video Dataset for Fair Deepfake Detection

In response to the increasing sophistication of face-swapping technologies, we present a large-scale, high-resolution video dataset aimed at advancing research in deepfake detection. This collection comprises more than 188,000 synthetic videos and 22,067 authentic samples, carefully constructed to ensure demographic diversity across gender and ethnicity, and to reflect real-world audiovisual variability.The synthetic content was generated using a reproducible face-swapping pipeline based on a state-of-the-art identity-preserving method. Source videos were curated from multiple domains to ensure both diversity and visual coherence. In addition, the dataset includes richly structured metadata, enabling detailed and fair evaluation across multiple dimensions.This benchmark addresses key limitations found in existing resources, particularly those related to demographic imbalance and lack of scalability, and serves as a robust foundation for the development, benchmarking, and comparison of deepfake detection systems in a transparent and generalizable manner.