we present VidGen-1M, a superior training dataset for text-to-video models. Produced through a coarse-to-fine curation strategy, this dataset guarantees high-quality videos and detailed captions with
excellent temporal consistency.We open source the VidGen-1M dataset so that scholars can train their own models and conduct fair model evaluation。
Due to network and size limitations, we split the dataset into 2048 parts and upload them one by… See the full description on the dataset page:
https://huggingface.co/datasets/Fudan-FUXI/VIDGEN-1M.
