gaunernst/flux-dev-portrait

该数据集使用[FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev)模型在默认设置（引导比例=3.5，50步）下生成，分辨率为1024x1024。提示词取自[k-mktr/improved-flux-prompts-photoreal-portrait](https://huggingface.co/datasets/k-mktr/improved-flux-prompts-photoreal-portrait)。图像被转换为uint8并使用无损WEBP编码，打包为WebDataset格式。该数据集可用于模型的蒸馏或自我蒸馏，以训练FLUX的新能力。目前生成的面孔缺乏多样性，未来的版本可能会通过不同的提示词增加多样性。

The FLUX-dev portrait dataset is generated using the FLUX.1-dev model with default settings (guidance scale = 3.5, 50 steps) at 1024x1024 resolution. The prompts are sourced from the k-mktr/improved-flux-prompts-photoreal-portrait dataset. The generated images are converted to uint8 format and encoded with lossless WEBP, packed in WebDataset format. This dataset can be used for distillation to another model or self-distillation when training new capabilities for FLUX. Currently, the generated face images lack diversity, and future versions may add more diversity with different prompts.