How to remove an image background

Updated 2026 · ~4 min read · by kizura

Removing a background turns a photo into a clean cutout you can place on any color, banner, or product page. Modern AI makes this almost automatic. Here's how it works and how to get the cleanest result.

How AI background removal works

An AI segmentation model looks at an image and predicts which pixels belong to the main subject and which belong to the background. It produces a mask, and the background pixels are made transparent. Older methods needed manual selection or a green screen; today a model trained on millions of images can separate a subject in seconds. Our background remover runs this model directly in your browser, so your image is never uploaded.

Getting the cleanest cutout

1. Use good contrast. A subject that stands out clearly from its background gives the model the easiest job.
2. Prefer even lighting. Harsh shadows can blend into the subject and confuse the edge detection.
3. Start from the highest resolution you have. More detail means more accurate edges.
4. Mind the hair and fur. Fine strands are the hardest part; a clean, slightly blurred background behind hair helps a lot.

What to do with the cutout

Once you have a transparent PNG, you can drop it onto a solid color for a product photo, composite it into a banner, or place it over a new scene. Our tool also lets you export with a white or black background instead of transparency, which is handy for marketplaces that require a plain backdrop.

Common problems

Edges look rough. Try a higher-resolution source or an image with better subject/background contrast.

Part of the subject is missing. Very low contrast between subject and background can cause this; a different photo usually fixes it.

It's slow the first time. The AI model (~40MB) downloads once on first use, then is cached for future runs.

FAQ

Is it really free and private?

Yes. Processing happens in your browser; the image never leaves your device and there is no cost.

What format is the result?

A PNG with a transparent background, which preserves the clean edges of your cutout.

Open the background remover to try it.

How one click finds the edge of a subject

Modern background removal is salient-object segmentation: a neural network trained on hundreds of thousands of hand-masked photos predicts, for every pixel, the probability that it belongs to the foreground subject. The output is not a binary stencil but an alpha matte — a soft transparency map — which is what lets single hairs and semi-transparent edges survive instead of being guillotined. ImgScale runs this model (an RMBG-family network) entirely in your browser: the model file downloads once, is cached, and your image never leaves your device for this tool.

The four classically hard cases — and what helps

• Fly-away hair and fur: the matte handles most of it; what kills it is a busy background the same colour as the hair. Shoot or choose images with tonal contrast behind the subject.
• Glass, smoke, veils: semi-transparency is genuinely ambiguous — expect a judgement call, and prefer solid replacement backgrounds where the softness reads naturally.
• Subject touching same-coloured background (white shirt on white wall): give the model any edge to work with — even slight shadow separation transforms results.
• Multiple subjects: saliency models keep what looks “main”. If it kept the wrong thing, crop toward your subject first and re-run.

After the cutout: formats and finishing

Export choices matter more here than anywhere else. Transparency only survives in PNG or WebP — export a cutout to JPG and the alpha channel is flattened to a solid colour permanently (we measured the same 800×800 logo: 5.9 KB as a transparent PNG, 22.6 KB as a white-boxed JPG — bigger and broken). For product work, composite onto pure white for marketplace listings (specific requirements here); for design work keep the transparent PNG as your master and let each destination flatten its own copy. If the edge shows a faint halo of the old background, a 1-pixel contraction or a subtle drop shadow in the editor hides the seam far better than re-running the model ever will.