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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add thinking (BAGEL-Think) to BAGEL | BAGEL · PhyEditBench normal data | Overall score: 6.09 -> 6.43 | Improved |
| add thinking (BAGEL-Think) to BAGEL | BAGEL · PhyEditBench anti-physics | Overall score: 5.11 -> 6.16 | Improved |
| use PhyWorld (test-time scaling + latent reduction) instead of Frame2Frame video-based editing | video-based editing · PhyEditBench normal data | Overall score: 5.38 -> 6.43 | Improved |
| use PhyWorld instead of Frame2Frame video-based editing | video-based editing · PhyEditBench anti-physics | Overall score: 3.81 -> 6.39 | Improved |
| evaluate ChronoEdit-14B on anti-physics instead of normal data | ChronoEdit-14B · PhyEditBench | Overall score: 8.51 -> 4.99 | Hurt |
| evaluate Seedream4.0 on anti-physics instead of normal data | Seedream4.0 · PhyEditBench | Overall score: 8.47 -> 6.95 | Hurt |
| evaluate PhyWorld on anti-physics instead of normal data | PhyWorld · PhyEditBench | Overall score: 6.43 -> 6.39 | No clear effect |
| move from single-step TypeA to global TypeE editing | BAGEL · PhyEditBench normal data | Type score: 6.01 -> 4.99 | Hurt |
| move from single-step TypeA to global TypeE editing | Step1X-Edit · PhyEditBench normal data | Type score: 7.16 -> 6.15 | Hurt |
| move from single-step TypeA to global TypeE editing | PhyWorld · PhyEditBench normal data | Type score: 6.59 -> 6.36 | No clear effect |
| use PhyWorld instead of ChronoEdit-14B | video-based editing · PhyEditBench normal data | Overall score: 8.51 -> 6.43 | Hurt |
| use PhyWorld instead of Gemini-2.5 on anti-physics | anti-physics editing · PhyEditBench anti-physics | Overall score: 7.07 vs 6.39 | Mixed |
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.
- OmniGen2 — OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026) · better on 2 of 2 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
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.88)
- Envisioning Beyond the Pixels: Benchmarking Reasoning-Informed Visual Editing (NeurIPS 2025 · similarity 0.88)
- PhyS-EdiT: Physics-aware Semantic Image Editing with Text Description (CVPR 2025 · similarity 0.87)
- KRIS-Bench: Benchmarking Next-Level Intelligent Image Editing Models (NeurIPS 2025 · similarity 0.87)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.87)
- Reasoning to Edit: Hypothetical Instruction-Based Image Editing with Visual Reasoning (ICML 2026 · similarity 0.87)
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.
