Research map / Image generation / Fast diffusion sampling
BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion
Xuan Ju, Xian Liu, Xintao Wang, Yuxuan Bian, Ying Shan, Qiang Xu
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 |
|---|---|---|---|
| use dual-branch design instead of single-branch SD Inpainting (ablation) | SD Inpainting · BrushBench | ImageReward (x10): 11.00 -> 11.59 | Improved |
| fine-tune the base UNet in the dual-branch design (ablation) | BrushNet · BrushBench | PSNR: 19.86 -> 20.13 | Improved |
| use VAE encoder instead of random conv layers for masked image (ablation) | BrushNet · BrushBench | PSNR: 14.89 -> 19.86 | Improved |
| include mask in branch input (ablation) | BrushNet · BrushBench | PSNR: 17.96 -> 19.86 | Improved |
| remove text cross-attention from the extra branch (ablation) | BrushNet · BrushBench | LPIPS (x10^2, lower better): 19.44 -> 16.09 | Improved |
| add full UNet feature layer-by-layer instead of half (ablation) | BrushNet · BrushBench | PSNR: 19.01 -> 19.86 | Improved |
| use full UNet feature addition instead of ControlNet-style addition (ablation) | BrushNet · BrushBench | LPIPS (x10^2, lower better): 24.09 -> 16.09 | Improved |
| add blurred-mask blending in pixel space (ablation) | BrushNet · BrushBench | PSNR: 19.86 -> 29.88 | Improved |
| blurred blending instead of direct paste (ablation) | BrushNet · BrushBench | CLIP Sim: 26.80 -> 26.81 | Mixed |
| add blending operation to BrushNet at inference | BrushNet · BrushBench outside-inpainting | CLIP Sim: 27.33 -> 27.22 | Hurt |
| replace ControlNet-Inpainting with BrushNet | ControlNet-Inpainting · BrushBench inside-inpainting | PSNR: 12.39 -> 21.65 | Improved |
Later papers that compared with this method
16 later accepted papers report BrushNet as a baseline in their results tables; 15 of them report a better number on most of the results they share with it.
- SONIC: Spectral Optimization of Noise for Inpainting with Consistency (ECCV 2026 · better on 3 of 3 shared results)
- From Inpainting to Layer Decomposition: Repurposing Generative Inpainting Models for Image Layer Decomposition (CVPR 2026 · better on 2 of 2 shared results)
- SketchAssist: A Practical Assistant for Semantic Edits and Precise Local Redrawing (CVPR 2026 · better on 2 of 2 shared results)
- Token Painter: Training-Free Text-Guided Image Inpainting via Mask Autoregressive Models (AAAI 2026 · better on 1 of 2 shared results)
- Accelerated Likelihood Maximization for Diffusion-based Versatile Content Generation (ECCV 2026 · better on 2 of 2 shared results)
- MTADiffusion: Mask Text Alignment Diffusion Model for Object Inpainting (CVPR 2025 · better on 11 of 13 shared results)
- One Stone with Two Birds: A Null-Text-Null Frequency-Aware Diffusion Models for Text-Guided Image Inpainting (NeurIPS 2025 · better on 11 of 11 shared results)
- Anywhere: A Multi-Agent Framework for User-Guided, Reliable, and Diverse Foreground-Conditioned Image Generation (AAAI 2025 · better on 9 of 9 shared results)
8 more not listed.
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
- RAD: Region-Aware Diffusion Models for Image Inpainting (CVPR 2025 · similarity 0.89)
- Paint by Inpaint: Learning to Add Image Objects by Removing Them First (CVPR 2025 · similarity 0.89)
- TurboFill: Adapting Few-step Text-to-image Model for Fast Image Inpainting (CVPR 2025 · similarity 0.89)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.89)
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models (ICLR 2025 · similarity 0.88)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 2024 · similarity 0.88)
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.
