HDAV

We introduce HDAV (High-definition Audio-Visual dataset), a large-scale dataset for one-shot human video synthesis. HDAV contains 2,203 identities with synchronized audio-visual content, comprising 3,425 web-collected and 1,325 curated videos. After preprocessing and manual annotation (including gender, clothing, and motion attributes), we obtain 2,173 high-quality video sequences with diverse motions and high clarity. To support 3D-consistent and multi-view modeling, HDAV further includes 360 multi-view videos from 12 angles across 30 identities. All videos are processed for consistent frame rate and human region cropping. The dataset, including processed sequences, annotations, and raw links, will be released for research use.