This dataset contains AI generated video frames created by different algorithms, namely: CogVideo, LumaAI (NeRF), Pika, Sora, Stable Video Diffusion (SVD), VideoCrafter v1 and VideoCrafter v2. Additionally, to train a synthetic video detector, we have included real video frames from the VideoASID and Moment-in-Time datasets. Note that video frames from the Moment-in-Time dataset are not included in this repository due to licensing restrictions. Therefore, to recreate our experiments listed in the paper below, you will need to download the Moment-in-Time dataset from the official website.
Our dataset's license is CC BY-NC-SA 4.0. Citing information can be found in the citation section.
For more information about dataset composition or other details, please checkout the
CVPR WMF 2024 paper.
