What it does IOPaint is a browser-based image editor that masks out bits of photos and fills the gaps using AI. It handles three jobs: erase (remove objects/defects), replace (swap in something new via diffusion), and out-paint (extend the canvas). You run it locally with iopaint start, point your browser at localhost:8080, and start scribbling masks.

The interesting bit The project doesn’t lock you into one model. It hot-swaps between specialized erasers like LaMa and a dozen-odd Stable Diffusion inpainting variants (SDXL, BrushNet, PowerPaintV2, etc.), plus plugs in Segment Anything for auto-masking, RealESRGAN for upscaling, and GFPGAN for face repair. It’s basically a modular pipeline where the UI stays constant and the backends are interchangeable.

Key highlights

• Fully self-hosted, runs on CPU, GPU, or Apple Silicon; one-click Windows installer available
• Batch processing via CLI: feed it folders of images and masks, get folders of results back
• Plugin ecosystem includes interactive segmentation, background removal, anime-specific segmentation, super-resolution, and two face-restoration models
• Supports loading local .ckpt/.safetensors files alongside HuggingFace downloads
• Companion macOS/iOS app (OptiClean) for quick object erasure on Apple devices

Caveats

• AMD GPU support on Linux only; Windows ROCm explicitly not supported by PyTorch yet
• Backend requires a manual restart after Python code changes during development
• Model downloads happen at first startup, which can mean a long initial wait

Verdict Worth a look if you want a local, non-subscription alternative to cloud inpainting tools and don’t mind managing PyTorch installs. Skip it if you need a polished consumer app—this is a tinkerer’s workbench with a web UI.