OpenVid-1M 提供文本-视频对，可把描述拆成主体、动作、镜头和环境四类字段，再归纳成自己的提示词模板。数据集卡明确写有研究/非商业限制，适合学习和内部实验，不应把样本视频当作自由商用素材。 OpenVid-1M provides text-video pairs that can be analyzed into subject, action, camera, and environment fields to build prompt templates. Its dataset card states research/non-commercial limits, so use it for learning and internal experiments rather than treating samples as free commercial stock.