Which mode should I use?

Selfie / portrait for photos of people — it's a model trained specifically for human subjects. General for anything else: products, pets, objects, or scenes.

Why is the first run slower?

The AI model (a few megabytes) has to download and initialize the first time you use a given mode in a session. Your browser caches it, so it's faster on subsequent uses.

What image formats are supported?

PNG and JPG input, with a transparent PNG as output.

Why didn't it cleanly separate my subject?

Automatic background removal works best with a clear subject against a reasonably distinct background. Busy scenes, low contrast between subject and background, or motion blur can all make the edges less precise.

Is this the same quality as a paid background-removal service?

It uses real, published open-source AI models — quality is generally good for everyday use, though a service using a larger, proprietary model may edge it out on the hardest cases (fine hair, transparent or reflective objects).

Does a high-resolution photo get a proportionally sharper mask?

Not in General mode — its AI model always predicts the subject mask at a fixed internal resolution, then scales it back up to your photo's real size, so extremely fine edges like wisps of hair can soften slightly in the upscale regardless of your original resolution.

Does Selfie mode need anything special from my browser?

It runs through WebGL via TensorFlow.js, so it needs a browser and device with WebGL support. That's the case for the vast majority of modern browsers, but it won't work if WebGL is disabled or unavailable.

What happens if processing fails partway through?

You'll see an on-screen message suggesting a different photo or the other mode — there's no partial or corrupted download, since the result only appears after processing finishes successfully.

Background Remover

About the Background Remover

Automatically isolates the subject of a photo and makes everything behind it transparent, using an on-device AI model that runs entirely in your browser — the image is never uploaded anywhere.

100% Free Runs in Your Browser No Sign-Up Required

How to use it

Upload a PNG or JPG photo.
Choose Photo type — General for products, objects, or pets, or Selfie / portrait for photos of people.
Click Remove background. The first run downloads the model (a few MB); later runs are faster.
Download the result as a transparent PNG.

Formula

General mode runs U^2-Net (Qin et al., "U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection," Pattern Recognition, 2020), a neural network trained to predict which pixels belong to the main subject of an image, producing a soft mask that's applied as the output's alpha channel. Selfie/portrait mode instead runs Google's MediaPipe Selfie Segmentation model, optimized specifically for people rather than general objects. Both run as ONNX/TensorFlow.js models entirely client-side via WebAssembly and WebGL — no server, no upload.

Worked example

A product photo shot against a plain studio backdrop typically comes out cleanly with General mode; a headshot or selfie usually segments more precisely in Selfie / portrait mode.

Recommendations

• Use Selfie / portrait mode for photos of people — it's trained specifically for that case and tends to produce cleaner edges around hair and shoulders than General mode.
• Use General mode for products, pets, or other objects — Selfie mode won't detect a non-human subject.
• A clear, reasonably distinct subject against its background gives the cleanest result; busy or low-contrast scenes are harder for any automated tool to isolate precisely.

Frequently asked questions

No. Both modes run as AI models loaded into your browser and executed locally — the photo itself never leaves your device.

Related tools