This repository provides custom nodes for ComfyUI to enhance image inpainting and outpainting capabilities. It targets users of Stable Diffusion XL and ComfyUI who require more control and advanced techniques for filling or expanding image regions, offering pre-processing and post-processing steps for seamless results.

How It Works

The nodes leverage several techniques for inpainting: Fooocus inpainting models for SDXL, LaMa, and MAT for pre-filling masked areas, and specialized nodes for mask expansion and blurring. A key feature is the VAE Encode & Inpaint Conditioning node, which efficiently combines existing image content with inpainting models, avoiding redundant VAE encoding for improved performance. Pre-processing options include neutral, Telea, and Navier-Stokes fill methods for different content generation needs.

Quick Start & Requirements

• Install via ComfyUI Manager or by cloning the repository into ComfyUI/custom_nodes.
• Requires opencv-python for Telea and Navier-Stokes fill modes (pip install opencv-python).
• Fooocus inpainting models, LaMa, and MAT models need to be downloaded and placed in ComfyUI/models/inpaint.
• Official workflows are available in the repository.

Highlighted Details

• Supports Fooocus inpainting models for SDXL, allowing transformation of standard checkpoints into inpaint models.
• Includes pre-processing nodes: Expand Mask, Fill Masked (neutral, Telea, Navier-Stokes), and Blur Masked.
• Integrates LaMa and MAT inpainting models for object removal and outpainting.
• Offers post-processing node Denoise to Compositing Mask for fine-grained control over blending based on denoise strength.

Maintenance & Community

• Project acknowledges contributions from Fooocus, LaMa, and MAT.
• Installation instructions point to ComfyUI Manager, indicating potential community integration.

Licensing & Compatibility

• No explicit license is mentioned in the README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

Distilled SDXL checkpoints (Turbo, Lightning, Hyper) are not compatible with the Fooocus inpainting method. The README does not specify a license, which may impact commercial adoption.