Free · runs in your browser · no upload

Video background remover

Remove the background from a video of a person, free. Drop in a clip, check one frame, then export the person on a transparent background, a solid colour or a blurred version of the original, with the sound kept. No watermark, and the video never leaves your computer.

Your file is read by this browser tab and never sent anywhere. The first run downloads the AI model from Hugging Face and caches it, so later runs start faster and work offline.

How the video background remover works

  1. Matting. Each frame goes through MODNet, a model that outputs a soft mask: how much of every pixel belongs to the person. Soft rather than on/off, so hair and edges blend instead of looking cut out with scissors.
  2. Smoothing. Each mask is blended a little with the previous frame's, which calms the flicker you get when every frame is cut out on its own.
  3. New background. The person is placed over transparency, the colour you pick, or a heavily blurred copy of the original frame, which looks like a shallow-focus camera.
  4. Encoding. The browser's own video encoder writes the result, and the original sound track is carried across.

Which background to choose

  • Transparent for compositing: put a presenter over slides, a product shot or other footage in your editor.
  • Solid colour for a clean look, or green (the default) if your editor's chroma key is what you are used to.
  • Blurred original to hide a messy room or a busy street while keeping the light and colour of the real scene.

Getting a clean result

The model does best when the person is clearly separate from what is behind them: some distance from the wall, even light and a background that is not the same colour as their clothes or hair. Fast movement blurs the edges of the person in the original frames, and that blur shows up as a soft halo. Check a frame with Preview before processing the whole clip.

When the desktop app is the better tool

Background removal works on one clip you already have open. Finding that clip is the slow part when your footage is spread over drives and folders. The Reelary desktop app indexes your whole library on your own computer, so "presenter in front of the whiteboard" or "wide shot of the stage" finds the right take by describing it.

Its editor then assembles the clips you find into a cut, and exports an MP4 or a timeline for Premiere, Final Cut or DaVinci Resolve, where you can finish compositing.

The desktop app finds the shots you want across your whole footage library by describing them, so the clip you cut out starts in the right place. Reelary is free to download for Mac and Windows; indexing and search have no file limit.

Questions

How do I remove the background from a video for free?

Drop the video into the tool above, press Preview frame to check the cut-out, choose transparent, a colour or blur, and press Remove background. You get the video back with the same sound and no watermark. Nothing is uploaded and there is no account.

Can I get a video with a transparent background?

Yes. Choose Transparent and the tool writes a WebM file with an alpha channel (VP9). Chrome, Edge, Firefox and OBS show the transparency. Editing apps differ in whether they import WebM with transparency, so check that yours does before relying on it. MP4 with H.264 cannot hold transparency at all, which is why the transparent export is WebM.

Does it work on objects, products or animals?

No. The model is trained to separate people from their surroundings: talking heads, webcam recordings, dancers, presenters. On a video without a person it will cut out little or nothing. Press Preview frame first to see what you will get.

Is the video uploaded?

No. Your browser decodes the video, the MODNet model runs in this tab on your graphics card (WebGPU) or CPU, and the new video is encoded by the browser. The model is downloaded once from Hugging Face and cached.

What are the limits?

Up to 30 seconds per run, at up to 1280 pixels on the long side and 30 frames per second; set a start and end to choose which 30 seconds. Every frame goes through the model: with WebGPU a 30-second clip takes roughly one to four minutes on a recent laptop (transparent export is the slowest), while on the CPU each frame takes one to two seconds, so keep clips short there. Hair, fast motion and a background close in colour to the person can leave soft or flickering edges.

Which browsers does it work in?

Recent Chrome and Edge on a desktop computer work best. It needs WebCodecs to encode the result, which Safari and Firefox have in recent versions too, but the transparent WebM export has only been tested in Chrome and Edge.

Which model does it use?

MODNet, a portrait matting model released under the Apache 2.0 license, run with Transformers.js. The finished video is encoded with WebCodecs through the Mediabunny library.

More free tools

Published October 19, 2026 · Models we use