Nothing is uploaded
The model runs inside your browser tab. Your image never touches a server — ours or anyone’s.
Media utilities
Remove the background from any image, right in your browser. No upload, no account, no credits, no watermark.
What it does
AI segmentation that runs on your machine, with the outputs the job actually needs.
The model runs inside your browser tab. Your image never touches a server — ours or anyone’s.
Keep true transparency, or swap the background instantly without re-processing.
No credits, no per-image pricing, no watermark, no account. Paste and go.
How to use it
In depth
With remove.bg being sunset, a lot of people need a new home for a very routine task. This tool covers the core of that workflow — image in, clean cutout out, transparent PNG down — with two differences worth knowing before you switch.
The first difference is where the work happens. remove.bg uploaded every image to its servers; this tool runs the open-source U²-Net segmentation model inside your browser through WebAssembly. That is why it is free without limits — there is no server bill to meter — and why it is safe for images you would not send to a third party: unreleased products, customer screenshots, ID photos.
The second difference is honest expectations. remove.bg ran large proprietary models; this runs a compact open model on your machine. It does very well on the common cases — people, products, and logos that stand apart from their background — and less well on wisps of hair, fur, and cluttered scenes. For a thumbnail, a slide, a profile photo, or a product shot, the result is typically ready as-is.
Around video, cutouts do their best work as overlays: a presenter freed from their webcam rectangle, a product dropped onto a branded thumbnail, a face on a chapter card. Those are exactly the images Mindstamp places on video as hotspots, buttons, and images — clickable, and reported per viewer.
Questions
For the everyday workflow, yes: image in, background removed, transparent PNG out, free. It runs in your browser instead of on a server, so there are no credits, accounts, or upload steps — and no server-side quality tier either.
No. The AI model downloads to your browser once and runs locally. The image, the mask, and the result never leave your machine.
Yes. Because processing happens on your device, each image costs nothing to anyone — so there is nothing to meter, watermark, or upsell.
It runs U²-Net, a well-known open-source segmentation model, through ONNX Runtime compiled to WebAssembly. The model predicts which pixels are the subject, and the mask becomes the alpha channel of your PNG.
The first use downloads the model and runtime (about 16 MB), which your browser then caches. After that, most images process in a few seconds.
People, products, and logos that stand apart from the background. Very fine hair, fur, glass, and busy scenes are harder for any compact model — check the edges before publishing those.
Related tools
These tools share engines and hand off to each other.
Design a completion certificate in your browser, download it as a print-ready PNG: or paste a class list and get one per learner, zipped.
Open the tool →Describe your product and get a scene-by-scene explainer script: narration, visuals, on-screen text, and one clear call to action.
Open the tool →Paste a transcript and see exactly where your video could ask, branch, link, and capture: moment by moment.
Open the tool →With Mindstamp
Mindstamp places images on video as clickable hotspots, buttons, and overlays — and reports every viewer who clicks.