VIDS-Guard Dataset (v1.0, Part 1 of 2)

VIDS-Guard Dataset (v1.0) is a comprehensive deepfake video dataset developed to support research in video-based forgery detection and digital forensics. The collection contains 26,975 labeled videos (13,487 real and 13,488 fake) sourced from eleven publicly available benchmarks including CelebV-HQ, DeeperForensics-1.0, DFDC, and FaceForensics++. All videos are stored in .mp4 format and organized into “Real” and “Fake” directories. This dataset provides high-quality, full-length videos representing diverse manipulation techniques (GANs, face-swap, reenactment) and real-world conditions such as varying illumination, ethnicity, and pose. It is specifically designed for benchmarking machine-learning and deep-learning models for deepfake detection. Access is restricted to academic, non-commercial research users due to licensing constraints of included sources. Note: This record is Part 1 of 2 (≈48 GB). Please also download Part 2 — VIDS-Guard Dataset (v1.0, External Test Set) under DOI [insert Part 2 DOI here] to obtain the full collection. See the attached README for dataset structure, citations, and license details.