Image Background Remover

Continue in the tool

More pages for this tool

remove background hair edges clean cutout

Remove a photo background without ruining hair and edges

Every automatic cutout fails in the same places: loose hair, clothing outlines and areas where the subject and background have similar colors. This page shows which control fixes each symptom.

Symptom, cause and the control that fixes it

The neural network sees the photo at low resolution: 512 px on the shorter side in fast mode and 512 x 512 px in high quality mode. The mask it returns has that same resolution and, when scaled up to the photo, is soft at the outlines. The edge snap (guided filter) uses the original photo's luminance as a guide and makes the mask outline obey the image's real edges, which brings hair strands back and sharpens the shoulder line. At 0 it is off; the default is 50.
Leftover background as a haze, like a pale smoke around the subject, is fixed with Clean up leftovers: it raises the alpha cut-off up to a maximum of 90%, erasing what the model marked with little certainty. A light or dark outline of 1 or 2 pixels calls for Shrink with a negative value; an outline cut too deep calls for a positive value. These values are in pixels on a photo with a 1000 px longer side and scale with the image: 3 equals 3 px at 1000 px, 6 px at 2000 px and 12 px at 4000 px.
When no global control is enough, the brush steps in. In Keep mode you paint over what the model erased by mistake; in Erase mode, over what is left of the background. The stroke is soft, has an adjustable size and Ctrl+Z undoes up to 25 steps. The original photo shows faintly behind the cutout so you can see what you are restoring. Glass, smoke, fine fur against a similar background and transparent objects remain hard for any model; in those cases the brush is the way, not one more slider.

Which adjustment to use

Hair strands disappeared

Input: Ajustar a borda à foto
Expected output: 50 para 80 ou 100

The guided filter brings back detail the mask blurred.

Pale haze around it

Input: Limpar resíduos
Expected output: 20 a 50

At 100 the cut-off reaches 90% alpha.

A 2 px light outline on a 2000 px photo

Input: Encolher
Expected output: -1 (equivale a 2 px)

Radius = value x longer side / 1000.

Shrink -3 on photos of different sizes

Input: 1000 px / 2000 px / 4000 px
Expected output: 3 px / 6 px / 12 px

The visual effect is the same at any resolution.

Part of an arm erased by the model

Input: Pincel Manter
Expected output: pinte sobre a área

With the original photo faintly shown as a guide.

Full tool FAQ

Is my photo uploaded to any server?

No. The photo is read and processed in your browser, inside a Web Worker. The internet is only used to download the cutout model (from Hugging Face) and the ONNX Runtime engine (from jsDelivr), once each. You can verify it in your browser's Network tab: after the model files are downloaded, no request carries your image.

What is downloaded the first time, and how big is it?

What is the difference between fast and high quality mode?

Why does high quality use a 512 px version instead of the 1024 px one?

Does it work on a phone?

Which file formats and sizes are accepted?

How do I get a white background for Mercado Livre or Amazon listings?

The cutout got a part wrong. How do I fix it?

Is the photo's resolution kept?

Can I use the cutouts commercially? What about the model licenses?

Frequently asked questions

Which model is better for hair?

BiRefNet lite (high quality) tends to deliver better edges and strands than MODNet, which is built for portraits and is simpler. In either, the guided edge snap improves the result, and fast mode is worth trying first because it weighs 6.6 MB.

Does the brush change the original photo?

No. The brush only changes the cutout mask, and only for that photo. The original file is never modified and you can clear all touch-ups with one button.

Why do Shrink values depend on the photo size?

Because 2 px of outline on a small photo is a lot and on a 4000 px photo it is almost nothing. The value is in pixels of a 1000 px photo and the tool scales it to the real resolution, so the same number gives the same visual result on any photo.

Next steps

Symptom, cause and the control that fixes it

The neural network sees the photo at low resolution: 512 px on the shorter side in fast mode and 512 x 512 px in high quality mode. The mask it returns has that same resolution and, when scaled up to the photo, is soft at the outlines. The edge snap (guided filter) uses the original photo's luminance as a guide and makes the mask outline obey the image's real edges, which brings hair strands back and sharpens the shoulder line. At 0 it is off; the default is 50.

Leftover background as a haze, like a pale smoke around the subject, is fixed with Clean up leftovers: it raises the alpha cut-off up to a maximum of 90%, erasing what the model marked with little certainty. A light or dark outline of 1 or 2 pixels calls for Shrink with a negative value; an outline cut too deep calls for a positive value. These values are in pixels on a photo with a 1000 px longer side and scale with the image: 3 equals 3 px at 1000 px, 6 px at 2000 px and 12 px at 4000 px.

When no global control is enough, the brush steps in. In Keep mode you paint over what the model erased by mistake; in Erase mode, over what is left of the background. The stroke is soft, has an adjustable size and Ctrl+Z undoes up to 25 steps. The original photo shows faintly behind the cutout so you can see what you are restoring. Glass, smoke, fine fur against a similar background and transparent objects remain hard for any model; in those cases the brush is the way, not one more slider.

Which adjustment to use

Hair strands disappeared

Input: Ajustar a borda à foto
Expected output: 50 para 80 ou 100

The guided filter brings back detail the mask blurred.

Pale haze around it

Input: Limpar resíduos
Expected output: 20 a 50

At 100 the cut-off reaches 90% alpha.

A 2 px light outline on a 2000 px photo

Input: Encolher
Expected output: -1 (equivale a 2 px)

Radius = value x longer side / 1000.

Shrink -3 on photos of different sizes

Input: 1000 px / 2000 px / 4000 px
Expected output: 3 px / 6 px / 12 px

The visual effect is the same at any resolution.

Part of an arm erased by the model

Input: Pincel Manter
Expected output: pinte sobre a área

With the original photo faintly shown as a guide.

Full tool FAQ

Is my photo uploaded to any server?

What is downloaded the first time, and how big is it?

What is the difference between fast and high quality mode?

Why does high quality use a 512 px version instead of the 1024 px one?

Does it work on a phone?

Which file formats and sizes are accepted?

How do I get a white background for Mercado Livre or Amazon listings?

The cutout got a part wrong. How do I fix it?

Is the photo's resolution kept?

Can I use the cutouts commercially? What about the model licenses?

Frequently asked questions

Which model is better for hair?

Does the brush change the original photo?

No. The brush only changes the cutout mask, and only for that photo. The original file is never modified and you can clear all touch-ups with one button.