A minimal prompt-pair dataset for spectacles bias studies in text-to-image diffusion.
base_prompt: the neutral prompt
with_spectacles_prompt: the same prompt with "wearing spectacles"
pair_index: stable pairing id
train: 800 examples
test: 200 examples
Measure latent deltas (with − without) on UNet outputs
Learn/text-side debias directions; cross-attention analyses
License… See the full description on the dataset page: https://huggingface.co/datasets/shirsho12/spectacles-bias-prompts. 
