floyo logo
Powered by
ThinkDiffusion
⚡️Nano Banana 2 ⚡️ just landed. Start creating now.
floyo logo
Powered by
ThinkDiffusion
⚡️Nano Banana 2 ⚡️ just landed. Start creating now.

ComfyUI_DiffuEraser

251

Last updated
2025-12-02

DiffuEraser is a specialized diffusion model designed for video inpainting, integrated within the ComfyUI framework. It enables users to effectively remove unwanted elements from video frames while maintaining visual coherence.

  • Supports high-resolution video inpainting up to 1280x720 with only 12GB of VRAM.
  • Incorporates advanced masking techniques to refine output quality and minimize artifacts.
  • Offers two blending modes in the sampling node to enhance flexibility in output results.

Context

DiffuEraser serves as an extension for ComfyUI, focusing on the task of video inpainting. This tool allows users to intelligently fill in or replace areas of video content, making it particularly useful for tasks like removing watermarks or unwanted objects seamlessly.

Key Features & Benefits

One of the standout features of DiffuEraser is its ability to utilize various diffusion models, including ProPainter and BrushNet, which enhance the inpainting process. The tool also allows for the adjustment of mask dilation parameters, enabling users to fine-tune the extent of the masking effect, which is crucial for achieving high-quality results.

Advanced Functionalities

DiffuEraser includes advanced capabilities such as dual-output blending options, which allow users to choose between reducing flicker in the output or utilizing a composite method to avoid repetitive model loading. This flexibility can significantly improve the workflow for users who require consistent and high-quality video outputs.

Practical Benefits

By integrating DiffuEraser into their workflow, users can expect improved control over video editing tasks, resulting in higher quality inpainting results. The tool enhances overall efficiency by streamlining the video editing process, allowing for quicker iterations and adjustments without compromising on output fidelity.

Credits/Acknowledgments

The development of DiffuEraser is credited to Xiaowen Li and collaborators, with references to various foundational models and techniques cited in the documentation. The project is open-source, allowing for community contributions and further enhancements under the specified license.

Inner Nodes

DiffuEraser_Loader, DiffuEraser_PreData, DiffuEraser_Sampler, Propainter_Loader, Propainter_Sampler