Civiverse 6M

该数据集名为Civiverse prompt，包含了数百万张图片及相关元数据，专注于提示分析，特别是对文本提示的语义特征进行研究。该数据集为每张图片提供了正面和负面的提示，元数据包括图片URL、图片哈希、帖子ID、时间戳、使用唯一哈希匿名化的用户名，以及生成过程中使用的各种参数。该数据集的规模达到了6,546,165张图片。这项任务的目的是分析文本到图像生成过程中用户参与度及提示特性的分析。

This dataset, named Civiverse Prompt, contains millions of images and their associated metadata, focusing on prompt analysis, particularly research into the semantic features of text prompts. For each image, the dataset provides both positive and negative prompts. The accompanying metadata includes image URLs, image hashes, post IDs, timestamps, usernames anonymized using unique hashes, and various parameters employed during the image generation process. The dataset has a total of 6,546,165 images. The primary goal of this work is to analyze user engagement and prompt characteristics in text-to-image generation processes.