This repository contains the training dataset for Pusa-V1.0, a video generation model that surpasses Wan-I2V with only a fraction of the training cost and data. The dataset features 3,860 high-quality video-caption pairs from Vbench2.0, originally generated by Wan-T2V-14B. By fine-tuning the state-of-the-art… See the full description on the dataset page: https://huggingface.co/datasets/RaphaelLiu/PusaV1_training.