Remove an Image Background and Refine the Edges

How do I remove a background cleanly?

Open Remove background, run the local model, then inspect the mask instead of treating the first result as final. Use Erase to remove background pixels the model retained and Restore to bring back subject pixels it removed. Fine hair, fur, glass, shadows, motion blur, and low-contrast boundaries often need manual correction.

The image stays in your browser. On first use, the tool downloads an AI model of about 220 MB from Hugging Face plus processing resources. The app uses a pinned revision of the BEN2 ONNX model repository. That repository’s description and licence apply to the model; a successful download does not guarantee a correct mask for a particular image.

Refine the mask before styling the result

1. Add a still JPG, PNG, or WebP image and choose Remove background. The shared limits are 30 images, 20 MB per image, 100 MB combined, 24 megapixels per image, and 80 megapixels combined, with an 8192-pixel maximum side.
2. Wait for the local model and the first mask. Cancellation keeps images that already completed.
3. Zoom in around the subject. Use a smaller correction brush near hair, fingers, product handles, and other narrow shapes. Erase only confirmed background; Restore only confirmed subject.
4. Check the edge on both light and dark preview backgrounds. A halo that disappears on one colour may remain obvious on another.
5. Repeat the review for every batch image. A correction on one composition does not prove that another subject was segmented correctly.

Choose transparency or a composed background

Keep a transparent output when the destination supports it, normally as PNG or WebP. JPG cannot preserve transparency and needs a solid background. The editor can also place the cutout on a solid colour or selected background image, add a shadow, trim empty space, and position or scale the subject.

For an exact canvas, set its width and height, then use subject padding, scale, and position. Padding creates space around the detected subject inside that composition; it is different from the Resize images “Fit with padding” mode, which fits a whole rectangular image inside exact output dimensions.

Edge problems and practical corrections

• Light or dark fringe: inspect the mask at high zoom on contrasting backgrounds and erase the narrow retained fringe.
• Hair or fur disappears: Restore conservatively with a small, softer brush. Some source detail may be too blended with the background to recover automatically.
• Transparent or reflective object is incomplete: compare continuously with the original. Glass and reflections may share the background’s colour and cannot be inferred reliably.
• Shadow looks cut off: decide whether the source shadow belongs to the subject. Restore it where appropriate or add a new composition shadow.
• Edges look good in preview but poor after download: inspect the encoded output at 100% and confirm the chosen format, dimensions, background, and quality.

Final export checks

Check the whole perimeter, interior gaps, transparent pixels, canvas dimensions, and subject placement. Test the downloaded image in its actual destination. Background removal changes visible pixels; it does not establish ownership, authenticity, or permission to use the subject.

For format and transparency choices, read JPG vs PNG vs WebP vs AVIF. For resizing the completed cutout, see Compress vs resize images.