Research map / Image generation / Editing and style transfer
KV-Edit: Training-Free Image Editing for Precise Background Preservation
Tianrui Zhu, Shiyi Zhang, Jiawei Shao, Yansong Tang
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 |
|---|---|---|---|
| cache and reuse background key-value pairs instead of regenerating background | KV-Edit on FLUX.1-dev · PIE-Bench | PSNR: 35.87 vs 32.53 (FLUX Fill), 24.44 (RF Edit) | Improved |
| add no-skip inversion step (NS) (ablation) | KV-Edit object removal · PIE-Bench object removal | Image Reward vs source prompt (lower is better): 6.87 -> 3.17 | Improved |
| add reinitialization (RI) to NS (ablation) | KV-Edit object removal · PIE-Bench object removal | Image Reward vs source prompt (lower is better): 3.17 -> 0.22 | Improved |
| add attention mask during inversion to NS+RI (ablation) | KV-Edit object removal · PIE-Bench object removal | Image Reward vs source prompt (lower is better): 0.22 -> 0.90 | Hurt |
| add NS, RI and AM together (ablation) | KV-Edit object removal · PIE-Bench object removal | HPS x10^2: 26.76 -> 26.51 | Hurt |
| add reinitialization and no-skip (+NS+RI) (ablation) | KV-Edit · PIE-Bench | PSNR: PSNR 35.87 -> 33.30, CLIP Sim 22.39 -> 23.62 | Mixed |
| use inversion-free variant of KV-Edit (ablation) | KV-Edit · 512x512, 28 steps | KV cache memory: 19.4G -> 1.9G | Improved |
| use inversion-free variant of KV-Edit | KV-Edit · PIE-Bench | artifacts | Hurt |
Later papers that compared with this method
2 later accepted papers report KV-Edit as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- T-Edit: Triple-Branch Diffusion Anchoring for Consistent Editing (ICML 2026 · better on 2 of 5 shared results)
- Follow-Your-Shape: Shape-Aware Image Editing via Trajectory-Guided Region Control (ICLR 2026 · better on 1 of 1 shared result)
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
- Improving Editability in Image Generation with Layer-wise Memory (CVPR 2025 · similarity 0.89)
- LayerVerse: Finding the Sweet Spot for KV-Injection in Training-Free Image Editing (ECCV 2026 · similarity 0.88)
- CacheEdit: Efficient Multi-round Image Editing via Adaptive Token-wise Reuse. (ICML 2026 · similarity 0.87)
- Accelerated Image Editing via Consistency-Aware Source Token Pruning (NeurIPS 2026 · similarity 0.87)
- Move and Act: Enhanced Object Manipulation and Background Integrity for Image Editing (AAAI 2025 · similarity 0.85)
- Balancing Preservation and Modification: A Region and Semantic Aware Metric for Instruction-Based Image Editing (ICML 2025 · similarity 0.85)
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.
