Research map / Image generation / Editing and style transfer
SwapAnything: Enabling Arbitrary Object Swapping in Personalized Image Editing
Jing Gu, Nanxuan Zhao, Wei Xiong, Qing Liu, Zhifei Zhang, He Zhang, Jianming Zhang, HyunJoon Jung, Yilin Wang, Xin Eric Wang
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 |
|---|---|---|---|
| Swap latent feature z plus attention variables within a mask (targeted variable swapping) and apply location adaptation to attention-based baselines | Photoswap, P2P, PnP, MasaCtrl · Single-object swapping | background preservation | Improved |
| Remove latent feature swap (keep only mask and attention variable swap) (ablation) | SwapAnything · Qualitative ablation | context preservation | Improved |
| Add style adaptation (Masked AdaIN) (ablation) | SwapAnything · Qualitative ablation | texture consistency | Improved |
| Add scale adaptation (mask shape guidance) (ablation) | SwapAnything · Qualitative ablation | shape alignment | Improved |
| Add all appearance adaptations (ablation) | SwapAnything · Qualitative ablation | connection to source image | Improved |
| Use a mask for swapping (ablation) | SwapAnything · Qualitative ablation | background preservation | Improved |
| Use a fixed mask through diffusion rather than a dynamically extracted one | SwapAnything · Design choice | performance | Improved |
| SwapAnything vs baselines in human evaluation of overall quality | SwapAnything · Human evaluation (overall quality) | preference %: 49.3 vs 27.8 (Photoswap); 71.1 vs 10.5 (CopyPaste) | Improved |
| SwapAnything vs BlipDiffusion on object gesture | SwapAnything · Human evaluation (object gesture) | preference %: 41.5 vs 39.6 | No clear effect |
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.
- Photoswap — PHOTOSWAP: Personalized Subject Swapping in Images (NeurIPS 2023) · better on 4 of 4 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
- InstantSwap: Fast Customized Concept Swapping across Sharp Shape Differences (ICLR 2025 · similarity 0.89)
- Energy-Guided Optimization for Personalized Image Editing with Pretrained Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.86)
- Zero-shot Image Editing with Reference Imitation (NeurIPS 2024 · similarity 0.86)
- An Item Is Worth a Prompt: Versatile Image Editing with Disentangled Control (AAAI 2025 · similarity 0.86)
- AttnDreamBooth: Towards Text-Aligned Personalized Text-to-Image Generation (NeurIPS 2024 · similarity 0.85)
- Improving Editability in Image Generation with Layer-wise Memory (CVPR 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.
