Erase Unwanted Objects, People, and Text with Browser-Based MI-GAN AI
A stray tourist walking through the background of a vacation portrait, an overhead power line cutting across architectural photography, or a dust spot on an e-commerce product table can ruin an otherwise great shot. Traditional clone-stamp retouching requires manual texture matching, while most cloud-based "magic erasers" upload your personal photos to remote servers, shrink their resolution, or lock full-size downloads behind a subscription.
Our AI Object Remover runs the open-source MI-GAN (Mask-Aware Image Inpainting Generative Adversarial Network) pipeline locally inside your web browser using ONNX Runtime Web (WASM). Instead of downscaling your entire photo, the engine crops a localized context window around the area you paint, reconstructs the missing background texture up to 1024 px per pass, and blends the generated pixels back into your full-resolution canvas using a dual-pass separable box-mean feather filter—leaving every untouched pixel 100% identical to your original file.
Key Controls in the Magic Eraser Workspace
Once your image opens in the editor, you can combine precision masking tools with zoom and export controls across the top toolbar and right sidebar:
- Brush (🖌), Mask eraser (◌), and Move (✋): Paint over the unwanted element with the translucent crimson mask using Brush, trim accidental strokes with Mask eraser, or switch to Move (or hold the
Space bar) to pan across zoomed-in areas.
- Brush size (6 to 200 px) & Expand selection (0 to 15): Adjust your stroke diameter with the Brush size slider (or press
[ and ] on your keyboard). The Expand selection slider (set to 3 by default) automatically dilates your painted mask slightly before inference so soft cast shadows and object halos are erased cleanly.
- Zoom, Pan, and Original Comparison: Use Zoom in (
＋), Zoom out (−), Fit, 1:1, or Ctrl + Mouse Wheel to inspect fine edges up to 3200%. Multi-step regional history lets you click Undo (Ctrl+Z) or Redo (Ctrl+Y), and holding down the 👁 button reveals your untouched original photo for an instant before-and-after check.
- Flexible Output Format & Quality: Export your cleaned image as a lossless PNG, or switch the Format menu to JPG or WebP and fine-tune the Quality slider from
60% to 100% (default 92%).
How to Remove Unwanted Elements from a Photo Step by Step
- Load your photo: Click the upload box, drag and drop a
PNG, JPG, or WebP file (up to 25 MB), or paste an image directly from your clipboard with Ctrl+V.
- Paint over the distraction: Select Brush, adjust Brush size so it slightly overlaps the edges of the object, and paint over the person, sign, wire, or blemish you want to erase. If you make a mistake, use Mask eraser or click Clear selection.
- Run the AI inpainting pass: Click Remove. On your very first visit, your browser downloads the ONNX WebAssembly runtime and the
28 MB MI-GAN model once and caches them locally in your browser's Cache Storage for instant offline-speed reuse on future visits.
- Refine in small passes if needed: For large or complex objects crossing multiple background textures, erase them in two or three smaller passes rather than painting one massive mask across the whole image.
- Export your retouched image: Choose your preferred file format (PNG, JPG, or WebP) and click Download image to save the full-resolution result without watermarks.
Complete Your Photo Retouching Workflow on Images Mint
Pair the AI Object Remover with our other client-side editing utilities to prepare your visuals for publishing or e-commerce:
- Upscale low-resolution photos after cleaning: If you erased blemishes or unwanted text from an older, small image, boost its resolution 2× or 4× using our AI Image Upscaler.
- Frame and resize your composition: Trim outer distractions with the Image Cropper or scale the retouched photo to exact web dimensions with our Online Image Resizer.
- Add fresh typography or brand protection: After erasing old dates, signs, or clutter, place clean captions using the Add Text to Image tool or stamp your own logo over product shots with the Image Watermark tool.
- Optimize file weight for fast page loads: Shrink your exported file size while keeping visual clarity with our JPG Compressor or WebP Converter.
Frequently Asked Questions
Are my photos uploaded to a cloud server when I use this object remover?
No. Both the ONNX Runtime Web engine and the MI-GAN neural network execute 100% locally inside your browser using WebAssembly in a background Web Worker. Your private photos never leave your computer or phone.
Why does the first removal take a few extra seconds compared to subsequent edits?
During your first session, your browser downloads the 28 MB migan_pipeline_v2.onnx model file and stores it in your browser's local Cache API (imagesmint-objremover-models-v1). Every removal after that loads the model directly from local disk cache in milliseconds.
Does the AI reduce the resolution of my original photo?
No. The tool only processes the bounding box surrounding your painted brush strokes (plus a 60% surrounding context margin) and blends the filled pixels back into your original canvas. Your exported image keeps its exact native pixel width and height—supporting up to 30 megapixels on desktop browsers and 12 megapixels on mobile devices.
What is the best way to remove a large object or a person standing across two different backgrounds?
Instead of painting a single huge mask over the entire figure at once, erase the object in smaller sections—for example, remove the head against the sky first, then remove the torso against the trees, and finally clean up the shadow on the ground. Smaller targeted passes give the MI-GAN generator cleaner surrounding textures to reference.