metadata

task_categories:
  - text-to-image
pretty_name: FLUX-dev portrait

FLUX-dev portrait

This dataset is generated with FLUX.1-dev using default settings (guidance scale = 3.5, 50 steps) at 1024x1024 resolution. The prompts are taken from k-mktr/improved-flux-prompts-photoreal-portrait. The images are converted to uint8 and encoded with lossless WEBP, and packed in WebDataset format.

This can be used for distillation to another model or self-distillation when training new capabilities for FLUX.

Currently the generated faces are not diverse. Future versions may add more diversity with different prompts.