A benchmark dataset of synthetic full-body photographs with paired inpainting masks, for evaluating adversarial protections against unauthorized image-editing attacks.
BodyBench contains 648 synthetic images of clothed adult subjects, generated with FLUX.1-Dev across a controlled grid of 27 prompted nationalities, two prompted genders, and four everyday scenarios (park, street, kitchen, classroom). Each image is paired with 13 inpainting masks… See the full description on the dataset page:
https://huggingface.co/datasets/BodyBench/BodyBenchData.
