DropletVideo-10M

The challenge of spatiotemporal consistency has long existed in the field of video generation. We have released the open-source dataset DropletVideo-10M —the world's largest video generation dataset with spatiotemporal consistency. It contains 10 million high-quality video clips (totaling 20,400 hours) and redefines the training benchmarks for video generation technology from a data perspective.

视频生成领域长期面临时空一致性的技术挑战。我们发布了开源数据集DropletVideo-10M——全球规模最大的具备时空一致性的视频生成数据集。该数据集涵盖1000万条高质量视频片段，总时长达20400小时，并从数据视角重新定义了视频生成技术的训练基准。