BrushNet is Tencent ARC Lab’s open-source image inpainting model, released as the official implementation of the ECCV 2024 paper “BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion”. Given an image and a mask, BrushNet fills the missing region with content that matches the surrounding context and any text prompt, and it plugs into existing Stable Diffusion pipelines without retraining the base model. The repository has around 1,700 stars and was last updated in 2024.
Most inpainting approaches fine-tune the whole diffusion model or bolt on adapters that blur the boundary between kept and generated pixels. BrushNet instead adds a dedicated second branch that processes the masked image with dense per-pixel features, injected into the frozen base model layer by layer. The result is sharper preservation of unmasked content and more coherent fills, and the same weights work across different Stable Diffusion community models.
Free. BrushNet is released as open research code with pretrained weights; you run it on your own GPU at no cost.
Image-editing developers, diffusion-model researchers, and creators who need high-quality object removal or region replacement inside existing Stable Diffusion workflows.