FLUX.2-dev Synthetic 2M: a large-scale synthetic text-to-image dataset

About the dataset

FLUX.2-dev Synthetic 2M is a large-scale synthetic text-to-image dataset released on Hugging Face with a citable DOI, 10.57967/hf/8311. It is intended to support research on text-to-image generation, synthetic data, and generative models.

• About 2,282,665 synthetic image-caption pairs
• Images generated with FLUX.2-dev at 512x512 resolution (50 inference steps, guidance scale 3.5)
• Captions sourced unmodified from the Text2Image-2M dataset
• Packaged as WebDataset, 571 shards of about 4,000 samples each; every sample is a .png image, a .txt caption, and a .json metadata file recording the seed and generation parameters
• Generated through a distributed multi-GPU pipeline on the Kempner AI cluster

The dataset is released for research use and is subject to the upstream FLUX.2-dev and Text2Image-2M licenses. See the release note for samples.

Recommended citation: Khoshnevis, N., Guo, G., Vanden-Eijnden, E., Boffi, N., & Albergo, M. S. (2026). FLUX.2-dev Synthetic 2M [Data set]. Hugging Face. https://doi.org/10.57967/hf/8311
Download Paper