AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Do Text Edits Generalize to Visual Generation? Benchmarking Cross-Modal Knowledge Editing in UMMs

Xin Gao, Cheng Yang, Chufan Shi, Taylor Berg-Kirkpatrick

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Contrastive vision-language alignment · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add reasoning-augmented generation (edited model verbalizes edited fact first) (ablation)PMET on Ovis-U1 · UniKE OverallVQA accuracy (%): 9.71 -> 28.32Improved
add reasoning-augmented generation (ablation)MEMIT on Ovis-U1 · UniKE OverallVQA accuracy (%): 8.70 -> 24.41Improved
add reasoning-augmented generation (ablation)AlphaEdit on Ovis-U1 · UniKE OverallVQA accuracy (%): 7.42 -> 20.67Improved
add reasoning-augmented generation (ablation)PMET on BLIP3o-4B · UniKE OverallVQA accuracy (%): 18.51 -> 19.29Improved
add reasoning-augmented generation (ablation)AlphaEdit on BLIP3o-4B · UniKE OverallVQA accuracy (%): 16.12 -> 17.33Improved
add reasoning-augmented generation (ablation)AlphaEdit on OmniGen2 · UniKE OverallVQA accuracy (%): 11.50 -> 17.90Improved
add reasoning-augmented generation (ablation)MEMIT on OmniGen2 · UniKE OverallVQA accuracy (%): 8.73 -> 12.17Improved
use PMET instead of MEMITOvis-U1 direct · UniKE Overalltext-side efficacy (%): 59.84 -> 72.18Improved
use softened null-space AlphaEdit instead of PMETBLIP3o-4B reasoning-augmented · UniKE Overallreasoning accuracy (%): 53.20 -> 61.67Improved
move from canonical edit prompt (Stage 1) to rephrased prompt (Stage 2)attribute edits, averaged across models · UniKE attribute stagestext-side efficacy: reduces efficacy by roughly 70%Hurt
move from Stage 1 to later stagesattribute edits, reasoning-augmented · UniKE attribute stagesVQA accuracy: decreases by roughly 35--40%Hurt
compare PMET edit drift across models (conditioning pathway)Ovis-U1 vs BLIP3o-4B · conditioning pathway analysisrelative drift at DiT input: 0.077 vs 0.458 (BLIP3o-4B) and 0.435 (OmniGen2)Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy