Guide · Updated October 2026

How to Remove a Background from a Photo

Background removal cuts the main subject out of a photo and makes everything behind it transparent. You can do it free, in your browser, with no account: upload a photo, let the tool separate the subject, check the edges, and download a transparent PNG or WebP.

Quick start

1. Open the tool. Go to the Deback background remover. No account is needed.
2. Upload your photo. Drag it onto the page or tap to browse. JPEG, PNG, WebP, and HEIC files are accepted, up to 12 MB each.
3. Wait for the first-run download if it is your first visit. The tool downloads an AI model of about 44 MB into your browser, with progress shown. After that first download, the model is cached and later runs start without it.
4. Check the preview. The result appears against a checkerboard pattern. That pattern is how transparency is displayed; it is not part of the image.
5. Inspect the edges. Use the before/after compare slider to switch between the original and the cutout, and zoom in to check hair, fur, and small details.
6. Fix anything the automatic pass missed. Use the brush touch-up to paint back a missing piece or remove a patch of leftover background.
7. Pick a background and a format. Keep transparency, or choose white, a custom color, or a blurred background. Download as PNG or WebP to keep transparency.

What background removal actually does

Background removal is subject segmentation. An AI model examines every pixel of your photo and decides whether it belongs to the main subject — a person, a product, a pet — or to everything behind it. The background pixels are then made transparent, leaving a cutout you can place on any backdrop.

Transparency is stored in the image file itself. PNG and WebP both support an alpha channel, which records how transparent each pixel is. That is why a cutout downloaded as PNG or WebP can sit on a colored background in another program without a white box around it. JPEG has no alpha channel, so a JPEG version always has a solid background — white, in this tool.

Choosing a source image that works

Most problem cases are predictable, and most can be avoided before you upload.

• Contrast between subject and background helps. A dark subject on a light background, or the reverse, gives the model a clear boundary. A subject that matches its background in brightness and color — a brown dog on brown leaves, a white shirt against a white wall — is harder to separate.
• Sharp, well-lit photos produce cleaner edges. Motion blur and heavy noise make edges ambiguous. If the photo is out of focus, the cutout will be too.
• Use the original file, not a screenshot or a thumbnail. A 300-pixel image cannot produce a crisp large cutout. A screenshot of a photo also bakes in whatever was on screen. Upload the original photo file.
• Watch what sits directly behind the subject. Fences, foliage, crowds, and other objects that touch the subject's outline make the boundary harder to find. A plain or distant background is easier.
• Expect hair and fur to be the hard case. Fine strands, flyaway hairs, and fluffy fur are the most difficult detail in background removal. Some stray hairs may be lost in the automatic pass. That is a limitation of the technology, and the brush touch-up exists for the pieces worth recovering by hand.
• Only process images you have the right to edit. Removing a background does not change who owns the photo.

A worked example: a product photo

Suppose you sell handmade mugs and need a listing photo with no background. Start with a straight-on photo of one mug on a kitchen table, in daylight, filling most of the frame.

1. Upload the photo. The tool separates the mug from the table and the kitchen behind it.
2. Drag the before/after compare slider to confirm the whole mug is present, including the handle. Handles are a common miss because the hole through the handle looks like background.
3. Zoom in on the rim and the handle. If the background shows through the handle hole, that is correct — the hole is genuinely empty. If part of the handle itself is missing, paint it back with the brush touch-up.
4. Check the bottom edge where the mug meets the table. A soft shadow under the mug may survive as a gray patch. Remove it with the brush if you want a fully clean cutout, or keep it if the shadow helps the mug sit naturally.
5. Choose a transparent background and download as PNG. That file is your master copy: you can place it on white for one marketplace, on a color for another, and keep the transparency for later.

Checking your result

Do not judge a cutout from the small preview alone. Small views hide edge problems that show up at full size. Check three things before you download.

• Use the before/after compare slider. Drag it back and forth to compare the cutout against the original. This is the fastest way to spot a missing piece: a finger, an ear, a strap, or a handle that the automatic pass dropped.
• Zoom in on the edges. Use zoom in, zoom out, and reset to inspect the outline at full size. Pay attention to hair, fur, thin straps, jewelry, text on the subject, and any place where the subject meets a busy background.
• Look at the outline against different backgrounds. Switch between transparent, white, and a custom color. A faint light halo around a dark subject shows up clearly against a dark color; a dark fringe shows up against white. If the outline looks clean on both, it will hold up in most uses.

Fixing halos, missing pieces, and leftover background

Three problems account for most bad cutouts. All three are repairable with the brush touch-up, which paints areas back into the subject or removes them from it.

• Halos: a light or dark fringe around the subject. A halo is background color bleeding into the edge pixels. It happens where the subject's edge was soft or the background was bright. Zoom in, then paint over the fringe to remove it. Work in short strokes along the outline rather than scrubbing across it.
• Missing pieces: part of the subject was cut away. Handles, straps, ears, tails, and thin objects are the usual losses. The compare slider shows you what is missing against the original. Paint the missing piece back with the brush. If the piece is very thin or detailed, paint slightly past its edge and then clean up the outside.
• Leftover background: patches of the old background remain. Shadows under the subject, reflections, and background objects that touch the subject are the usual leftovers. Paint over them to remove them. If a shadow is large and soft, removing it completely can leave a hard edge; decide whether the shadow helps the subject sit naturally before removing all of it.

If the automatic result is wrong across most of the image — the subject itself is misidentified — the source photo is usually the problem. Try a different photo with clearer contrast rather than painting over a large area by hand.

Transparent or white background

The background options are transparent, white, custom color, and blurred. Which one to pick depends on where the image is going.

• Transparent keeps the cutout flexible. You can place it on any background later, in any program that supports transparency. Keep the transparent file as your master copy even if you also need a white version.
• White matches listing photos and documents that expect a plain background. A white background also hides small edge imperfections that would show on a colored backdrop.
• Custom color is for images that need a specific backdrop, such as a brand color or a social post background.
• Blurred keeps the original scene visible but out of focus behind the subject, which suits portraits and social photos.

Note that transparency only survives in PNG or WebP. If you choose a white, colored, or blurred background, the background is baked into the image, and JPEG becomes an option.

Choosing PNG, WebP, or JPEG

Two rules cover most cases. If you might need transparency later, download PNG or WebP now — you cannot recover transparency from a JPEG afterward. And never screenshot the preview: a screenshot bakes the checkerboard pattern into the picture. Use the download button.

Dimensions and file limits

Three limits apply. Knowing them avoids surprises.

• File size: 12 MB per file. Larger files are rejected before processing. If your photo is over the limit, export a smaller copy from your phone or photo program first.
• Output dimensions: your photo's dimensions, capped at 2048 pixels on the longest side. A 4000-pixel-wide photo becomes 2048 pixels wide; a 1200-pixel-wide photo stays 1200 pixels wide. Images are never enlarged, so a small upload stays small. This cap keeps processing within what a phone or laptop browser can handle.
• Extreme sizes are rejected before processing. Images over 12,000 pixels per side or over 100 megapixels are refused before any work starts.

For most uses — listings, profiles, social posts, documents — 2048 pixels on the longest side is more than enough.

Local processing and the first run

Processing happens on your device, in your browser, using the device's processor. Your photo is not uploaded to a server for processing. The only data sent from the tool is anonymous usage counts and, if you choose to leave one, a star rating. Nothing about the image itself leaves the device.

The first run in a browser downloads an AI model of about 44 MB, with progress shown while it downloads. After that first download, the model is cached in the browser, so later runs start without downloading again. Clearing the browser's stored data removes the cached model, and the next run downloads it again.

Two practical consequences follow. Keep the tab open while processing runs, especially on a phone — switching away can pause the work. And the first run is the slowest run in a given browser, because later runs skip the download.

Troubleshooting

Questions

Do I need an account?

No. There is no signup, no account, and no credit system. The limits are 12 MB per file, 50 photos per batch run, and the 2048-pixel output cap.

Is my photo uploaded anywhere?

No. Processing happens on your device. The only data sent is anonymous usage counts and an optional star rating.

Why is my output smaller than my original?

Outputs keep your photo's dimensions up to 2048 pixels on the longest side. A larger original is downscaled to that cap; a smaller original is never enlarged.

Why does the first run take longer?

The first run in a browser downloads an AI model of about 44 MB. After that, the model is cached and later runs start without the download.

Can I process many photos at once?

Yes. The batch tool handles 50 photos per run, with a ZIP download of all finished cutouts and a per-photo download and retry option for any photo that fails.

What if the automatic result is wrong across most of the image?

Try a different source photo with clearer contrast between subject and background. Hand-fixing a large area is slower than choosing a better photo.