Are these real results or marketing pictures?

Real. Every photo on this page was uploaded to this site like any other image, and the cutout is the file the server returned. The processing times shown under each pair are what the server actually took on the day they were made.

Why show failures at all?

Because you will meet them. Every background remover produces the same three failures, and a gallery of wins only teaches you that the tool photographs well. Knowing in advance that glass, fine hair and low contrast are the hard cases saves you the upload.

Can the hard cases be fixed by trying again?

Not by re-uploading the same file: the model is deterministic and will return the same cutout. What helps is changing the input, by shooting against a plain background that differs in brightness from the subject, avoiding backlight, and using a sharp file rather than a heavily compressed one.

How large were these images?

Source sizes are listed under each pair and range from 800 to 1600 px on the long side. Anything longer than 2200 px is scaled down before processing, so the results here are at full processed size.

What automatic background removal actually looks like

Twelve images pushed through this tool, unedited: 7 came out clean, 2 came out partly wrong and 3 failed outright. The failures are the useful part.

Everything below is a real result from this site, not a promotional render. Each caption gives the source size and the time the server took. The grey checkerboard is transparency: that is what you get in the PNG.

Cases that came out clean

Solid subjects, a background that differs from them, and an edge the model can find.

Original photo: mandarin on a plain backdrop — **Mandarin on a plain backdrop** The easy case, and worth showing as a baseline: a solid subject with a hard edge against an even backdrop leaves the model nothing to get wrong. 1600x1200 px source, 3.6 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: the easy case, and worth showing as a baseline: a solid subject with a hard edge against an even backdrop leaves the model nothing to get wrong. — **Mandarin on a plain backdrop** The easy case, and worth showing as a baseline: a solid subject with a hard edge against an even backdrop leaves the model nothing to get wrong. 1600x1200 px source, 3.6 s to process. Photo: Public domain, Wikimedia Commons.

Original photo: running shoe, studio light — **Running shoe, studio light** The loose lace is the part people check first, and it survived as a separate strand rather than being absorbed into the shoe. 1600x1600 px source, 3.5 s to process. Photo: CC0, Wikimedia Commons.

The same photo after background removal: the loose lace is the part people check first, and it survived as a separate strand rather than being absorbed into the shoe. — **Running shoe, studio light** The loose lace is the part people check first, and it survived as a separate strand rather than being absorbed into the shoe. 1600x1600 px source, 3.5 s to process. Photo: CC0, Wikimedia Commons.

Original photo: coffee pot and mug on a glass shelf — **Coffee pot and mug on a glass shelf** Two separate objects, a patterned glaze and a reflective shelf behind them. Both came out, including the gap inside the handle. 1299x1600 px source, 6.9 s to process. Photo: CC0, Wikimedia Commons.

The same photo after background removal: two separate objects, a patterned glaze and a reflective shelf behind them. both came out, including the gap inside the handle. — **Coffee pot and mug on a glass shelf** Two separate objects, a patterned glaze and a reflective shelf behind them. Both came out, including the gap inside the handle. 1299x1600 px source, 6.9 s to process. Photo: CC0, Wikimedia Commons.

Original photo: aloe in a pot, bright window behind — **Aloe in a pot, bright window behind** Backlight usually causes trouble, but the leaves are thick enough to hold an edge, and every spike stayed separate from the next. 1200x1600 px source, 4.7 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: backlight usually causes trouble, but the leaves are thick enough to hold an edge, and every spike stayed separate from the next. — **Aloe in a pot, bright window behind** Backlight usually causes trouble, but the leaves are thick enough to hold an edge, and every spike stayed separate from the next. 1200x1600 px source, 4.7 s to process. Photo: Public domain, Wikimedia Commons.

Original photo: guitarist in a street scene — **Guitarist in a street scene** A full figure in a cluttered outdoor scene, and the hardest part came out right: the gap between the arm and the guitar body is empty, not filled in. 1600x1200 px source, 3.5 s to process. Photo: CC0, Wikimedia Commons.

The same photo after background removal: a full figure in a cluttered outdoor scene, and the hardest part came out right: the gap between the arm and the guitar body is empty, not filled in. — **Guitarist in a street scene** A full figure in a cluttered outdoor scene, and the hardest part came out right: the gap between the arm and the guitar body is empty, not filled in. 1600x1200 px source, 3.5 s to process. Photo: CC0, Wikimedia Commons.

Original photo: child and a wet dog by a pool — **Child and a wet dog by a pool** Two subjects that are not touching, kept as one cutout, with the water and the poolside gone. Wet fur is easier than dry fur: it clumps. 800x531 px source, 3.7 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: two subjects that are not touching, kept as one cutout, with the water and the poolside gone. wet fur is easier than dry fur: it clumps. — **Child and a wet dog by a pool** Two subjects that are not touching, kept as one cutout, with the water and the poolside gone. Wet fur is easier than dry fur: it clumps. 800x531 px source, 3.7 s to process. Photo: Public domain, Wikimedia Commons.

Original photo: cat on a sofa, similar tones — **Cat on a sofa, similar tones** The body outline holds against a background of a similar brightness. Look closely at the whiskers: they are gone, and no automatic tool keeps them. 1600x1067 px source, 3.5 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: the body outline holds against a background of a similar brightness. look closely at the whiskers: they are gone, and no automatic tool keeps them. — **Cat on a sofa, similar tones** The body outline holds against a background of a similar brightness. Look closely at the whiskers: they are gone, and no automatic tool keeps them. 1600x1067 px source, 3.5 s to process. Photo: Public domain, Wikimedia Commons.

Cases that came out partly wrong

The subject is right and something extra came with it. The model decides what counts as foreground, and you cannot argue with it.

Original photo: carved wooden chair in a hallway — **Carved wooden chair in a hallway** The chair itself is cut well, including the openwork in the back. A pale strip of the doorframe behind it came along, because the model read it as part of the same object. 1200x1600 px source, 4.6 s to process. Photo: CC0, Wikimedia Commons.

The same photo after background removal: the chair itself is cut well, including the openwork in the back. a pale strip of the doorframe behind it came along, because the model read it as part of the same object. — **Carved wooden chair in a hallway** The chair itself is cut well, including the openwork in the back. A pale strip of the doorframe behind it came along, because the model read it as part of the same object. 1200x1600 px source, 4.6 s to process. Photo: CC0, Wikimedia Commons.

Original photo: three runners on a road — **Three runners on a road** The people are cut cleanly, but the traffic cone between them was kept as part of the subject. There is no way to tell the model otherwise: it decides what the foreground is. 1057x1570 px source, 4.7 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: the people are cut cleanly, but the traffic cone between them was kept as part of the subject. there is no way to tell the model otherwise: it decides what the foreground is. — **Three runners on a road** The people are cut cleanly, but the traffic cone between them was kept as part of the subject. There is no way to tell the model otherwise: it decides what the foreground is. 1057x1570 px source, 4.7 s to process. Photo: Public domain, Wikimedia Commons.

Cases that failed

Three failures that no automatic tool avoids. If your photo looks like one of these, save yourself the upload.

Original photo: wine glass against black — **Wine glass against black** The shape is right and the result is still wrong. Glass is see-through, and a cutout can only be opaque or absent, so the bowl comes back solid and the thing that made it glass is gone. 1067x1600 px source, 4.6 s to process. Photo: CC0, Wikimedia Commons.

The same photo after background removal: the shape is right and the result is still wrong. glass is see-through, and a cutout can only be opaque or absent, so the bowl comes back solid and the thing that made it glass is gone. — **Wine glass against black** The shape is right and the result is still wrong. Glass is see-through, and a cutout can only be opaque or absent, so the bowl comes back solid and the thing that made it glass is gone. 1067x1600 px source, 4.6 s to process. Photo: CC0, Wikimedia Commons.

Original photo: faded sepia studio card — **Faded sepia studio card** Almost no contrast between the sitter and the card she is printed on. The model found a face and threw away the rest, which is the worst kind of failure: confident and wrong. 1051x1600 px source, 3.5 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: almost no contrast between the sitter and the card she is printed on. the model found a face and threw away the rest, which is the worst kind of failure: confident and wrong. — **Faded sepia studio card** Almost no contrast between the sitter and the card she is printed on. The model found a face and threw away the rest, which is the worst kind of failure: confident and wrong. 1051x1600 px source, 3.5 s to process. Photo: Public domain, Wikimedia Commons.

Original photo: 1940s studio portrait, dark hair on a dark backdrop — **1940s studio portrait, dark hair on a dark backdrop** Hair that blends into the backdrop is the classic hard case. The face and dress are fine, and the hair comes back as a smear with a halo of the old background attached. 1200x1600 px source, 6.9 s to process. Photo: Public domain, Wikimedia Commons.

The same photo after background removal: hair that blends into the backdrop is the classic hard case. the face and dress are fine, and the hair comes back as a smear with a halo of the old background attached. — **1940s studio portrait, dark hair on a dark backdrop** Hair that blends into the backdrop is the classic hard case. The face and dress are fine, and the hair comes back as a smear with a halo of the old background attached. 1200x1600 px source, 6.9 s to process. Photo: Public domain, Wikimedia Commons.

How to pick a photo that will work

Put brightness between the subject and the background. Dark hair on a dark wall is the single most common failure. It is not about colour, it is about tone: a dark subject on a light backdrop is easy, the same subject on an equally dark backdrop is not.
Avoid backlight. A bright window behind the subject makes the edge glow and the model cuts into it. Light from the front or the side is worth more than any setting.
Use a sharp file, not a small compressed one. Heavy JPG compression smears the edge before the model ever sees it. A photo saved at low quality for messaging apps is much harder than the original.
Expect nothing from transparent material. Glass, veils, smoke and water keep their outline and lose their transparency. A cutout is opaque or absent, with nothing in between.
Do not expect individual hairs or whiskers. The outline of hair survives; strands do not. If you need strand level accuracy, budget a few minutes in an editor afterwards.

Questions people actually ask

Are these real results or marketing pictures?: Real. Every photo on this page was uploaded to this site like any other image, and the cutout is the file the server returned. The processing times shown under each pair are what the server actually took on the day they were made.
Why show failures at all?: Because you will meet them. Every background remover produces the same three failures, and a gallery of wins only teaches you that the tool photographs well. Knowing in advance that glass, fine hair and low contrast are the hard cases saves you the upload.
Can the hard cases be fixed by trying again?: Not by re-uploading the same file: the model is deterministic and will return the same cutout. What helps is changing the input, by shooting against a plain background that differs in brightness from the subject, avoiding backlight, and using a sharp file rather than a heavily compressed one.
How large were these images?: Source sizes are listed under each pair and range from 800 to 1600 px on the long side. Anything longer than 2200 px is scaled down before processing, so the results here are at full processed size.

Source photographs are public domain or CC0 files from Wikimedia Commons, linked under each pair. The cutouts were produced by this site on August 23, 2026, and the uploads were deleted afterwards.