Recently, developer scraed released LanPaint, an image repair tool that does not require additional training, on GitHub. The tool is designed to help users achieve high-quality image repair effects on any stable diffusion model (SD), even custom models trained by users themselves. LanPaint allows the model to "think" before denoising through multiple iterations, thus achieving more seamless and accurate repair results.
One of the main features of LanPaint is zero training fix. Users can use the tool immediately on any SD model without the tedious training process. In addition, LanPaint integration is very simple, and users can operate like using the standard ComfyUI KSampler. The smooth workflow greatly reduces the threshold for use.
In terms of functionality, LanPaint provides a high-quality seamless restoration experience. Users can use this tool to perform multiple types of repair tasks by simply following the instructions and dragging the image into ComfyUI. For example, the user can convert a basket image into a basketball image, or turn a white shirt into a blue shirt, etc. Different example results demonstrate LanPaint's powerful ability when processing complex images.
Using LanPaint is very simple. Users only need to install ComfyUI and ComfyUI-Manager and join the LanPaint node through search or manual installation. After installation, the LanPaint node appears in the ComfyUI's "Sampling" category, and users can perform high-quality image repairs like using the default KSampler.
During use, users need to note that LanPaint requires the use of a binary mask (value 0 or 1), and the transparency and hardness of the mask must be set to maximum to ensure compatibility. In addition, LanPaint is very dependent on user text prompts, and the user needs to clearly describe what they want to generate in the mask area.
LanPaint has brought revolutionary improvements to the field of image restoration, simplifying operational processes, improving repair quality, and providing users with more powerful image processing tools.
Project: https://github.com/scraed/LanPaint
Key points:
Zero training fix: Supports immediate use on any stable diffusion model without additional training.
Simple integration: The same workflow as standard ComfyUI KSampler, lowering the barrier to use.
High-quality repair: Provides high-quality, seamless image repair effects, supporting a variety of complex repair tasks.