AIML-TUDA/t2i-diversity-evalprompts

T2I多样性评估提示集是一个包含4340个英文提示的数据集，这些提示分为四个描述密度增量（从最小到最长）。基础提示来源于DrawBench和Parti-Prompts，每个不超过10个标记。对于每个基础提示，GPT-4o生成了保持主题不变的同时逐步增加细节的短、中、长三个变体，从而使得每个概念有4个变体，可以针对提示复杂性对文本到图像模型进行压力测试，而不改变底层场景。

T2I-Diversity Evaluation Prompt Set is a dataset containing 4340 English prompts organized into four increments of descriptive density (minimal → long). The base prompts, sourced from DrawBench and Parti-Prompts, are each ≤ 10 tokens. For each base prompt, GPT-4o generated short, medium, and long rewrites that keep the subject constant while progressively adding detail, resulting in 4 variants per concept for stress-testing text-to-image models across prompt complexity without changing the underlying scene.