AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

PhyEditBench: A Real-World Multi-Stage Benchmark for Physics-Aware Image Editing

Shengbin Guo, Shaokang He, Chaoyue Meng, Shengpeng Xiao, Xunzhi Xiang, Shaofeng Zhang, Qi Fan

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Multimodal LLM benchmarks · 2 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 thinking (BAGEL-Think) to BAGELBAGEL · PhyEditBench normal dataOverall score: 6.09 -> 6.43Improved
add thinking (BAGEL-Think) to BAGELBAGEL · PhyEditBench anti-physicsOverall score: 5.11 -> 6.16Improved
use PhyWorld (test-time scaling + latent reduction) instead of Frame2Frame video-based editingvideo-based editing · PhyEditBench normal dataOverall score: 5.38 -> 6.43Improved
use PhyWorld instead of Frame2Frame video-based editingvideo-based editing · PhyEditBench anti-physicsOverall score: 3.81 -> 6.39Improved
evaluate ChronoEdit-14B on anti-physics instead of normal dataChronoEdit-14B · PhyEditBenchOverall score: 8.51 -> 4.99Hurt
evaluate Seedream4.0 on anti-physics instead of normal dataSeedream4.0 · PhyEditBenchOverall score: 8.47 -> 6.95Hurt
evaluate PhyWorld on anti-physics instead of normal dataPhyWorld · PhyEditBenchOverall score: 6.43 -> 6.39No clear effect
move from single-step TypeA to global TypeE editingBAGEL · PhyEditBench normal dataType score: 6.01 -> 4.99Hurt
move from single-step TypeA to global TypeE editingStep1X-Edit · PhyEditBench normal dataType score: 7.16 -> 6.15Hurt
move from single-step TypeA to global TypeE editingPhyWorld · PhyEditBench normal dataType score: 6.59 -> 6.36No clear effect
use PhyWorld instead of ChronoEdit-14Bvideo-based editing · PhyEditBench normal dataOverall score: 8.51 -> 6.43Hurt
use PhyWorld instead of Gemini-2.5 on anti-physicsanti-physics editing · PhyEditBench anti-physicsOverall score: 7.07 vs 6.39Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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