Brief

Researchers have developed new benchmarks to evaluate the precise editing capabilities of visual AI models. PaintBench focuses on 20 fundamental image editing operations, finding that current industry leaders score only 17.1% on average. NRVBench, on the other hand, assesses non-rigid video editing, examining how well models can modify deformable motion while maintaining material-specific plausibility. Both benchmarks highlight significant limitations in current models' ability to perform complex, precise visual manipulations. AI