CaptionEmporium/midjourney-niji-1m-llavanext

midjourney-niji-1m-llavanext数据集包含2,079,886条合成描述，对应1,039,943张图片，这些图片来自midjourney-v6-520k-raw和nijijourney-v6-520k-raw数据集。描述是通过LLaVA-NeXt模型生成的，并经过清理和缩短处理。数据集主要用于文本到图像和图像到文本的任务，所有图片及其元数据以MozJPEG编码的JPEG格式存储在webdataset格式的`wds/`目录中。

The midjourney-niji-1m-llavanext dataset contains synthetic captions for 1,039,943 images sourced from midjourney-v6-520k-raw and nijijourney-v6-520k-raw. The captions are generated using tag generation with wd-swinv2-tagger-v3 and captioning with llama3-llava-next-8b, followed by cleanup and shortening with Meta-Llama-3-8B. The dataset includes 2,079,886 captions, all in English. The images are available as MozJPEG encoded JPEGs. The dataset is intended for tasks involving text-to-image and image-to-text generation.