MiniMax H3 – 1K is a model image and video test dataset designed to test the knowledge scope and generative capabilities of the MiniMax H3 model. This dataset contains 1,000 video clips, covering images of various sizes, as well as multiple media types and themes. The videos have a resolution of 768 pixels and were generated by minimax_h3_fl2va_pruned_int8_convrot.safetensors through 30 inference steps.