About this tool
Remove image backgrounds in your browser. No upload
Drop an image and the background is removed automatically, leaving a clean transparent PNG you can drop onto any design, listing, or document. The AI model runs on your own device, your photo never reaches a server. Works best on products and objects with clear edges; people come out well too, though very fine hair can be a little soft.
Why use OmegaPix
-
One click, transparent PNG : No tracing, no manual masking. Drop an image and get a cutout on a transparent background in a second or two. Download as PNG (with alpha) or WebP, or drop a solid colour behind it.
-
The AI runs on your device : Background removal is exactly the kind of task most tools send to a server. This one downloads a small (~4 MB) model once and runs it in your browser with WebGPU (or WASM where WebGPU is unavailable). Open DevTools → Network. Your image is never sent anywhere.
-
Free and unlimited : No account, no credits, no watermark, no per-image fee. Because the work happens on your machine, there is nothing to meter.
How it works
Drop your image
JPG, PNG, WebP, or AVIF. On first use the browser fetches the AI model once (~4 MB) and caches it, subsequent images are instant to start.
Background removed automatically
The model finds the main subject and cuts out everything else. You see the original, the mask, and the final cutout on a transparent checkerboard.
Choose output and download
Keep it transparent, or composite over white, black, or a custom colour. Download a PNG/WebP, or share it straight to another app on mobile. Files are named with a -no-bg suffix.
Frequently asked questions
Is my image uploaded to a server?
No. The AI model downloads to your browser once and then runs locally on your device. Your image is read with the File API, processed in a Web Worker, and never sent anywhere. You can confirm this in DevTools → Network: there is no request carrying your image bytes.
How good is the quality?
Very good on products, objects, and items with defined edges, clean, sharp cutouts. People come out well for body and clothing. The one honest limitation is very fine, wispy hair: the edge can be a little soft rather than strand-perfect. For most listings, social posts, and design work it is more than enough.
What formats can I download?
Transparent PNG (default) or WebP with an alpha channel. You can also composite the cutout over a solid white, black, or custom-colour background and download that.
Does it work on my phone?
Yes. It runs on modern mobile browsers (it uses WebGPU where available and falls back to WASM otherwise). The first image takes a moment longer while the model downloads and warms up; after that it is quick. You can share the result straight to another app from the share button.
Why does the first image take a few seconds?
The first run downloads the ~4 MB model and initialises the runtime. That happens once and is cached, so later images start immediately. Inference itself is typically well under a couple of seconds on a desktop.
Can I remove the background from several images at once?
V1 is single-image. Batch mode is a likely follow-up. For now, process images one at a time.
Can I touch up the edges by hand?
Not in V1. It is fully automatic. A manual refine brush (to restore or erase parts of the mask) is on the roadmap for tricky cases like flyaway hair.
Is it really free with no watermark?
Yes. No account, no credits, no watermark. The model runs on your device, so there is nothing to charge for per image.
When to use this tool
Product photos
Cut products out of busy backgrounds for marketplace listings, catalogues, and ads, clean white or transparent backgrounds in one step.
Profile & people shots
Drop yourself or a subject onto a solid colour or a new scene for avatars, team pages, and thumbnails.
Design assets
Extract objects, logos on photos, or icons to reuse as transparent PNGs in slides, posters, and mockups.
Stickers & cutouts
Make transparent cutouts for chat stickers, scrapbooks, and collages, then share them straight from your phone.
When not to use this tool
You need strand-perfect hair
For studio-grade hair matting (modelling portfolios, high-end retouching), a dedicated professional tool will do better on flyaway strands. This tool is for fast, everyday cutouts.
You need a layered/editable mask
The output is a flattened image with an alpha channel, not an editable selection in a layered editor. If you need to keep tweaking the mask, do the cutout in a full editor.
The subject is ambiguous
Scenes with no clear main subject (busy landscapes, crowds) are not what a salient-object model is built for; results will be unpredictable.
Technical details
The model
OmegaPix uses U2Netp. The lightweight (~4.4 MB) variant of U2-Net, a salient-object-detection network released under the Apache-2.0 licence. It predicts a mask of the main subject, which becomes the alpha channel of your image.
How it runs in the browser
The model is executed with onnxruntime-web inside a Web Worker so the page never freezes. It uses the WebGPU backend when your device supports it (fast, often sub-second) and falls back to WebAssembly otherwise. Both the model and the runtime are served from OmegaPix, no third-party CDN, no external request.
How the cutout is built
Your image is resized to the model input, normalised, and run through the network to produce a 320×320 saliency map. That map is min-max normalised and upscaled to the full image, then applied as the alpha channel of the original full-resolution pixels, so the output stays sharp.
Why on-device matters here
Background removal often involves personal photos, IDs, or unreleased product shots. Server-based removers see exactly those images. Running on your device removes that exposure entirely. The picture never leaves your machine.
Your files stay on your device
The image is read into your browser with the File API and processed by an AI model running locally in a Web Worker via onnxruntime-web. The model and runtime are downloaded once (and cached) from OmegaPix; your image itself is never uploaded and no third party sees it. Verify in DevTools → Network.