BG Remove Troubleshooting: When the Result Looks Wrong

The cutout edge looks like a staircase instead of a smooth line. This happens when the model uses a binary mask instead of alpha matting, or when the input image resolution is too low for the mask to be accurate.

Fix 1: process the image at higher resolution. If the original is 1000 pixels wide, the mask has 1000 pixels of resolution to work with. A 4000 pixel source gives the mask more detail.

Fix 2: use a model that supports matting. On Remova, BiRefNet Lite and ISNet both apply matting. U2-Net does not.

Fix 3: refine in the manual editor. The brush tool can smooth edges that the automatic process left rough.

The subject is preserved but fine hair strands are gone. The edge of the head is a hard line where hair should fade out.

This is a matting problem. The model detected the subject correctly but the alpha values in the transition zone were not computed.

Fix: reprocess with hair preservation enabled. On Remova, this is the default for human subjects. If the result still looks wrong, try a different model.

Some pixels inside the subject are transparent when they should be opaque. The result looks like the subject has holes.

This happens with images that have dark regions similar to the background, or with images that have shadows inside the subject.

Fix: use the restore brush in the manual editor to paint those regions back to opaque. For automated fixes, try a different model. BiRefNet is more conservative than U2-Net about what it declares foreground.

The tool kept the background and removed the subject, or it selected a different object in the frame.

This happens when the model misidentifies the subject. Common causes: the subject is small in the frame, the subject has low contrast with the background, or there are multiple candidate subjects.

Fix 1: crop the image so the intended subject fills more of the frame. Rerun.

Fix 2: use the manual editor to specify the region. Tools like Remova have a smart selection mode that lets you click the subject to define what should be kept.

Fix 3: reshoot with clearer subject-background separation.

The cutout looks mostly right but there are visible patches of background around the edges. Usually a halo of the original background colour.

This is a spill problem. The model declared the pixel as foreground but the pixel still contains colour from the background.

Fix: use the erase brush in the manual editor to remove the residual pixels. Or, for automated handling, look for a tool that applies spill suppression. On Remova, the manual mode has a spill suppression slider that removes the tint.

You expected a transparent PNG but got a JPEG with a white background. Somewhere in the pipeline the transparency was flattened.

Common causes: the input was a JPEG (JPEG cannot store transparency), the output format was set to JPEG instead of PNG, or the destination platform converted the image.

Fix: verify the output format. If the tool allows format selection, choose PNG. If you uploaded a JPEG, the tool will produce a PNG but the input JPEG may have already lost detail in the background.

The image has been processing for more than a minute. Something is wrong.

Common causes: the image is very large (over 20MB), the backend is under load, or the tool is processing the image at a higher resolution than necessary.

Fix: compress the image before uploading. Most tools process at a lower resolution than the source and composite back, so a compressed source produces the same result.

If the tool is slow consistently, check the tool's status. Free tools sometimes have slow periods. Remova typically processes images in fifteen to thirty seconds.