ViMix-14M. ViMix-14M is a curated multi-source video-text dataset of ~14 million pairs that provides crawl-free, download-ready access with high-quality, temporally-aligned captions, addressing the data bottleneck in open-source text-to-video generation models.