MTEdit-Bench
A benchmark for multi-turn image editing released alongside the MTEdit dataset, evaluating models across four dimensions: instruction following, content consistency, resistance to visual quality drift, and use of cross-turn context.
Timeline (last 90 days)
On MTEdit-Bench, Qwen-Image-MTEdit performs best among all evaluated open-source models.
MTEdit-Bench evaluates multi-turn editing capability along four dimensions: following editing instructions, maintaining content consistency, resisting visual quality drift, and effectively using cross-turn historical context.
All Facts (2)
On MTEdit-Bench, Qwen-Image-MTEdit performs best among all evaluated open-source models.
50%UnverifiedMTEdit-Bench evaluates multi-turn editing capability along four dimensions: following editing instructions, maintaining content consistency, resisting visual quality drift, and effectively using cross-turn historical context.
50%