Research map / Video / Video generation
ChronoEdit: Towards Temporal Reasoning for In-Context Image Editing and World Simulation
Jay Zhangjie Wu, Xuanchi Ren, Tianchang Shen, Tianshi Cao, Kai He, Yifan Lu, Ruiyuan Gao, Enze Xie, Shiyi Lan, Jose M. Alvarez, Jun Gao, Sanja Fidler and 2 more
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 temporal reasoning tokens at inference (Nr=10) (ablation) | ChronoEdit-14B · PBench-Edit | Overall (GPT-4.1): 4.43 -> 4.53 | Improved |
| add temporal reasoning tokens at inference (Nr=10) (ablation) | ChronoEdit-14B · PBench-Edit | Action Fidelity: 4.01 -> 4.31 | Improved |
| extend reasoning from 10 to 20 or 50 steps (ablation) | ChronoEdit-14B-Think · PBench-Edit | Overall (GPT-4.1): 4.53 -> 4.51 (Nr=20), 4.52 (Nr=50) | No clear effect |
| distill to 8-step Turbo student (ablation) | ChronoEdit-14B · ImgEdit Basic-Edit | Overall (GPT-4.1): 4.42 -> 4.13 | Mixed |
| scale from 2B to 14B | ChronoEdit · ImgEdit Basic-Edit | Overall (GPT-4.1): 4.13 -> 4.42 | Improved |
| scale from 2B to 14B with temporal reasoning | ChronoEdit-Think (Nr=10) · PBench-Edit | Overall (GPT-4.1): 4.44 -> 4.53 | Improved |
| initialize from pretrained video model instead of scratch (ablation) | ChronoEdit · image-pair training | training loss convergence | Improved |
| encode input with 4x repeated target jointly instead of individually (ablation) | ChronoEdit VAE encoding · 1000 video frame pairs | PSNR (dB): 40.21 -> 39.82 | No clear effect |
| video-prior editing model vs FLUX.1 Kontext [Dev] baseline | ChronoEdit-14B · ImgEdit Basic-Edit | Overall (GPT-4.1): 3.52 -> 4.42 | Improved |
Closest papers by meaning
- PhyEditBench: A Real-World Multi-Stage Benchmark for Physics-Aware Image Editing (ECCV 2026 · similarity 0.91)
- Learning Action and Reasoning-Centric Image Editing from Videos and Simulation (NeurIPS 2024 · similarity 0.89)
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.87)
- UniReal: Universal Image Generation and Editing via Learning Real-world Dynamics (CVPR 2025 · similarity 0.87)
- VisionDirector: Vision-Language Guided Closed-Loop Refinement for Generative Image Synthesis (CVPR 2026 · similarity 0.87)
- Pathways on the Image Manifold: Image Editing via Video Generation (CVPR 2025 · similarity 0.86)
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.
