1 paper across 1 session
We propose KRIS-Bench, a benchmark designed to evaluate knowledge-based reasoning in instruction-based image editing models.