Synthetic deep-sea mining videos, prompts, and evaluation data for four text-to-video systems

This dataset accompanies the manuscript “Prompt engineering for synthetic deep-sea mining video: A controlled evaluation of four text-to-video systems” (Ocean manuscript Ocean-2026-0015). The deposit contains 595 synthetic videos: 300 from an archived comparison of four configured text-to-video systems, 45 from an archived prompt word-order study, 10 exploratory image-to-video clips, and 240 from a controlled text-to-video extension. The extension covers four generation systems, four prompt conditions, five scenarios, and three requested seeds per experimental cell. The accompanying research materials include prompt records and video-to-prompt mappings, per-video scores for six VBench dimensions, available generation and evaluation records, analysis code, and de-identified ratings from three independent evaluators for a separately selected 60-video subset of the archived comparison. A manifest lists the video filenames, file sizes, and SHA-256 checksums. The file synthetic_videos_595.zip contains the MP4 videos. The file research_data_code_2026_09_24.zip contains the machine-readable research data and code. The manuscript and supplementary files are submitted separately to Ocean and are not included in this deposit. The archived image-to-video clips are exploratory and do not constitute a paired comparison with text-to-video. This dataset supports evaluation of visual characteristics and domain-rule compliance; it does not contain measured plume-physics results or downstream perception-task outcomes.