amazon-agi/Amazon-Nova-1.0-Reel-evals

该数据集包含用于评估Amazon Nova Reel模型的多样化提示集，这些提示涵盖了人类和活动、动物、自然风景和景观、室内场景、物体互动以及创意场景和活动等六个广泛类别。提示集设计用于评估模型在生成视频时对运动文本对齐、相机运动指令遵循、动态属性生成以及复杂协调运动生成的能力。整个提示集包含大约700个提示，均来自各种开源基准。

This dataset includes a diverse set of prompts designed to evaluate the Amazon Nova Reel model, covering six broad categories: human and activities, animals, natural scenery and landscapes, indoor scenes, objects interactions, and creative scenes and activities. The prompt set is structured to assess the models ability in motion-text alignment, following camera movement instructions, generating dynamic attributes, and producing complex, coordinated motions. The curated prompt set consists of approximately 700 prompts, all sourced from various open source benchmarks.