A MILLION-SCALE DATASET FOR TEXT-TO-VIDEO GENERATION

Summary from the TLDR AI newsletter

This dataset was curated to have minimal overlap with existing video datasets while including themes and actions that users training models might be interested in for final video synthesis and understanding. All videos are collected from the official YouTube creator API and are CC licensed.