Background Remover

Erase photo backgrounds without uploading

Loading…

Was this tool helpful?

About this tool

This background remover uses an AI model called U²-Net (u2netp) to erase everything behind the main subject in a photo, leaving a transparent background. The model runs directly inside your own browser rather than on a server, so the photo is never uploaded anywhere. Once you pick a photo and click run, it downloads the model (about 4.4MB) plus the engine that runs it in the browser (onnxruntime-web, about 13MB) — roughly 17MB combined — and feeds a 320x320 downscaled copy of the photo through it to work out what counts as background.

How to use

  1. Select the photo you want to remove the background from.
  2. Click run — this is when the AI model and its engine (about 17MB combined) are downloaded.
  3. Once the download finishes, the photo is downscaled to 320x320, analyzed, and the background is erased.
  4. Toggle the checker background on and off to see how cleanly the transparent area was cut.
  5. Download the result as a transparent-background PNG once you are happy with it.
  6. Click Other photo to try a different image. If you are still in the same tab, it processes right away since the model does not need to be downloaded again.
  7. Use Send below to pass the result straight to another tool (like compression) instead.

Frequently asked questions

Is my photo uploaded to a server?
No. Background removal runs a U²-Net AI model directly in your browser, so the photo data is never sent to a server.
Why does it need to download something the first time?
It needs the AI model (about 4.4MB) plus the engine that runs it in a browser (about 13MB), roughly 17MB total. To avoid forcing that download on people who never use this tool, it only downloads once you click "run." Within the same tab, later photos are processed instantly since the model is already loaded.
Is it free to use?
Yes. Both the model (U²-Net) and the runtime (onnxruntime-web) are distributed under open-source licenses that permit commercial use.
What if the result is not clean?
The photo is downscaled to 320x320 for analysis, so very fine edges like individual hair strands may not separate perfectly. Photos with clear contrast between subject and background tend to give cleaner results.
Why does it use a separate entry point instead of the default onnxruntime-web build?
The default entry point tries to download a build that includes WebGPU support (about 26MB), nearly doubling the size. Background removal only needs wasm, so this tool uses the smaller wasm entry point, about 13MB.
If I process several photos in the same tab, does it redownload 17MB every time?
No. The session is created once and reused, so from the second photo onward in the same tab, processing starts instantly with no further download. Opening a new tab or reloading the page requires downloading it again.
Can I send the result straight to another tool?
Yes. The Send button below the result lets you pass the file directly to another tool, such as compression, without downloading it and uploading it again.
Can I get the result in a format other than PNG?
No. Since the transparent background needs to be preserved, the result is always saved as a PNG with an alpha channel.

Analysis is done on a 320x320 downscaled copy of the photo, so very fine edges (like hair) may not separate perfectly. The computation also runs on your device's CPU, so large photos or slower devices can take a while to process.

Similar tools