IOPaint setup: object removal, inpainting and outpainting

This tutorial, uploaded on March 2, 2024, covers IOPaint, formerly Lama-Cleaner, for image editing on your own computer. The setup uses Python 3.10, PyTorch 2.2.1 and CUDA 12.1 on Linux. The source does not identify an IOPaint version; its installation steps and compatibility advice describe that period.

The presenter installs the app in a separate Conda environment, selects a GPU device and opens its browser interface on localhost port 8080. He explains automatic model downloads and reuse of single-file checkpoints. The walkthrough describes support for Stable Diffusion 1.5, SD2 and SDXL. He suggests around 8 GB of VRAM for most options, calls CPU processing slow, and expresses uncertainty about Mac hardware support.

Editing starts with brush masks and object removal using LaMa and ZITS. Selecting a diffusion model adds prompts and generation controls for replacement. A high-resolution image triggers a CUDA memory error; restricting the edit with the Cropper's default 512 x 512 area lets his example complete. Interactive segmentation helps him create another mask.

The outpainting demonstration expands the canvas with Extender. The presenter advises disabling histogram matching for this operation and shows that a prompt is optional. He finishes with PowerPaint's object-removal, shape-guided and text-guided modes. These examples show local AI image editing, with results that depend on the chosen model and mask quality.