Research map / Image generation / Fast diffusion sampling
UniEdit-Flow: Unleashing Inversion and Editing in the Era of Flow Models
Guanlong Jiao, Biqing Huang, Kuan-Chieh Jackson Wang, Renjie Liao
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 Uni-Inv instead of DDIM-like Euler inversion (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 74.90 -> 86.11 | Improved |
| add Uni-Inv instead of DDIM-like Euler inversion (ablation) | Uni-Edit · PIE-Bench editing (SD3) | structure distance (x10^3): 40.87 -> 21.40 | Improved |
| add Uni-Edit guidance instead of naive delayed injection (ablation) | Uni-Edit · PIE-Bench editing (SD3) | CLIP whole: 24.26 -> 26.39 | Improved |
| add Correction step (ablation) | Uni-Edit · PIE-Bench editing (SD3) | CLIP whole: 23.78 -> 26.39 | Improved |
| add Correction step (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 89.67 -> 86.11 | Hurt |
| use (1+m_i) rather than m_i as Correction mask weight (ablation) | Uni-Edit · PIE-Bench editing (SD3) | CLIP whole: 25.16 -> 26.39 | Improved |
| add velocity fusion with mask (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 85.50 -> 86.11 | Improved |
| add velocity fusion with mask (ablation) | Uni-Edit · PIE-Bench editing (SD3) | structure distance (x10^3): 22.92 -> 21.40 | Improved |
| add mask in Correction (region-adaptive) (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 83.13 -> 86.11 | Improved |
| add region-adaptive mask guidance instead of plain CFG (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 82.52 -> 86.11 | Improved |
| use inversion velocity at t_i instead of Uni-Inv (ablation) | Uni-Edit · PIE-Bench editing (SD3) | background SSIM: 80.25 -> 86.11 | Hurt |
| apply Uni-Inv to Heun sampler | Heun inversion (SD3) · Conceptual Captions reconstruction | PSNR unconditional: 16.98 -> 20.10 | Improved |
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.
- FireFlow — FireFlow: Fast Inversion of Rectified Flow for Image Semantic Editing (ICML 2025) · better on 6 of 6 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
- Unveil Inversion and Invariance in Flow Transformer for Versatile Image Editing (CVPR 2025 · similarity 0.90)
- Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations (ICLR 2025 · similarity 0.90)
- Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025 · similarity 0.89)
- Taming Rectified Flow for Inversion and Editing (ICML 2025 · similarity 0.89)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.89)
- SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing (NeurIPS 2025 · similarity 0.89)
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.
