ToolRobinNo account needed

Portrait Background Remover

Remove a portrait background locally with an on-device person segmentation model.

HEIC photos load an additional local decoder on first use; large originals can take longer to process.

Choose a clear photo with one visible person, start the local model and inspect the generated transparent PNG at full size before downloading.

Portraits only: the local model finds people, not products, logos or general objects. A first-use model download of about 11 MB is served by ToolRobin; your photo never leaves this device.

Choose a portrait. The local model loads only after you start.

How to use

Make a portrait PNG with a transparent background

Remove a portrait background locally with an on-device person segmentation model.

Step by step

  1. Provide the source using: Remove a portrait background.
  2. Choose: Remove background. Inspect the result before copying, downloading or using it elsewhere.
  3. Choose a clear photo with one visible person, start the local model and inspect the generated transparent PNG at full size before downloading.

Choosing the settings

OptionWhat changes
Remove a portrait backgroundUse a clear photo with one person near the camera and visible contrast from the background. The model is trained for portrait-style people segmentation, not merchandise, logos, signatures, passport-photo compliance or fine hair-level cutouts. Thin features, poor lighting, occlusion and small people can be missed; inspect the PNG edges before using it.
Remove backgroundThe first run downloads an approximately 11 MB segmentation engine plus a small model, then caches them in your browser. Your selected image is decoded and processed locally in a worker. Inputs are limited to 20 MB, 8 megapixels and 8192 pixels per side; the output is a PNG at the source dimensions, up to 40 MB. Processing stops if you cancel or leave the page.

Example and expected result

Dimension example: a supported 800 × 600 portrait produces an 800 × 600 PNG. The background becomes transparent where the model removes it; inspect hair and glasses. Matching dimensions does not prove that the mask is correct.

If it is not working

  • Use a clear photo with one person near the camera and visible contrast from the background. The model is trained for portrait-style people segmentation, not merchandise, logos, signatures, passport-photo compliance or fine hair-level cutouts. Thin features, poor lighting, occlusion and small people can be missed; inspect the PNG edges before using it.
  • The first model download needs a working connection even though your file is processed locally. If loading fails, retry the missing download; if memory is insufficient, use a smaller input. A model result still needs comparison with the original.
  • Inspect the exported file at its actual pixel size, especially text, edges and transparent areas. A preview that looks good when reduced can still have visible defects in the downloaded image.

Where portrait background removal works best

Use a clear photo with one person near the camera and visible contrast from the background. The model is trained for portrait-style people segmentation, not merchandise, logos, signatures, passport-photo compliance or fine hair-level cutouts. Thin features, poor lighting, occlusion and small people can be missed; inspect the PNG edges before using it.

Local model and limits

The first run downloads an approximately 11 MB segmentation engine plus a small model, then caches them in your browser. Your selected image is decoded and processed locally in a worker. Inputs are limited to 20 MB, 8 megapixels and 8192 pixels per side; the output is a PNG at the source dimensions, up to 40 MB. Processing stops if you cancel or leave the page.

You might also need