How to use Image Inpainting: input to export
Follow the complete workflow, export checks, and privacy guidance for Image Inpainting.
What image inpainting can handle
Image inpainting is useful for removing small objects, text, dust, scratches, and passers-by. It reconstructs the selected area from nearby texture, so grass, sky, walls, water, and repeating surfaces often work well. Faces, precise building lines, or subjects covering a large part of the image are less likely to finish cleanly in one pass.
Steps
- Upload a PNG, JPG, or WebP image. It is read inside the browser and is not uploaded.
- Adjust the brush, cover the item completely, and overlap its edge slightly.
- Use Erase Selection, Undo, or Redo to correct the mask.
- Start repair. The first run downloads an approximately 62 MB LaMa ONNX model that browsers normally cache.
- Drag the original comparison control to inspect the boundary, then download a PNG.
Why smaller passes are better
A very large mask leaves less surrounding context for reconstruction. When several objects are far apart, repair one, review the result, and then paint the next. Small passes also work better near image edges.
Privacy and device requirements
Images and masks are processed only in this browser and are not uploaded to FreeTool. The first repair downloads an approximately 62 MB LaMa model that browsers normally cache. Inference runs on the current device. Speed depends on its processor and available memory; use a desktop browser if a low-memory phone cannot finish.