Remove background from image — free, no sign-up, runs in your browser

The short version

1. Open the background remover.
2. First time only: wait ~10 seconds while the AI model downloads (~30 MB, cached forever after).
3. Drop your image.
4. Wait 2–5 seconds.
5. Download a PNG with transparent background.

The first visit is the slow one. Every visit after is instant — your browser caches the model. So this stays fast at volume.

Why this used to cost money

Background removal used to require a GPU on a server because the models were too heavy to run anywhere else. You'd upload an image, a cloud GPU would process it, you'd download the result. The pricing models (remove.bg, PhotoRoom, Canva Pro) reflected the cost of keeping those GPUs humming.

Two things changed:

1. The models got smaller. The BRIA and ISNet-class segmenters shipped as ~30 MB quantised ONNX files with accuracy close to the heavyweight counterparts.
2. WebAssembly + WebGPU caught up. ONNX Runtime Web and similar runtimes let the browser run the same model the cloud does — just on the user's own device.

Competitors still charging $9-$99/mo are mostly charging for convenience (no setup) and for their brand. The technical moat is gone. We didn't invent the model — we just wired it into a browser tab.

When our tool is equal to (or better than) a paid service

• People shots with a clear background separation — headshots, product photos on neutral backgrounds, social media selfies. Indistinguishable from remove.bg at HD resolution.
• Privacy-sensitive material — ID photos, confidential documents, medical images. Paid services vary in what their processing agreements say about your uploaded files. Ours doesn't have a processing agreement because there's nothing to process on our end.
• High volume — remove.bg's Pro tier caps you at 200 images/month for $9. We don't cap. (We literally can't — we never see the conversions.)
• Offline work — visit the tool once with internet, disconnect, keep working. The model runs locally.

When a paid service still wins

Being honest about this matters more than the marketing:

• Hair-level detail on complex backgrounds — if you're cutting out a subject with wispy hair against a cluttered background, remove.bg's premium model (what they call "HD") catches a few more strands than the 30 MB browser model. Worth paying if you edit portraits professionally every day.
• Batch processing tied into a design app — if you're already inside Canva or Photoshop all day, their built-in remover saves you a tab-switch. Our tool is a better deal for the occasional cutout, not for the designer living in Canva.
• API volume at scale — remove.bg has an API for devs who want to remove 10k backgrounds per day from a server. That's on our roadmap but not shipped yet.

Tips for best results

1. Start with a reasonably-sized image

The model runs best on images between 1–4 megapixels. A phone photo straight out of camera (12 MP+) works but takes longer and uses more RAM. If you're getting memory errors, resize to 1920 px wide first with our resize tool.

2. Cleaner background = sharper cutout

A subject on a plain wall beats a subject in front of a similar- coloured object. If the background and the foreground have very similar colours (brown coat on wood panelling, white cat on a white couch), the model will leak in a few pixels. That's the physics of segmentation, not a bug.

3. Download PNG, not JPEG

JPEG doesn't have an alpha channel. If you export the cutout as JPEG, the transparent pixels will show up as either black or white. PNG preserves transparency — that's why our tool always outputs PNG for this specific operation.

4. If you need a coloured background

Remove the background first, then use our image compressor with a hex-coded background colour to flatten the alpha onto any colour you like. Two-step flow, but gives you full control.

Honest comparison: where each tool lands

Common questions

Is this really free forever?

Yes. We don't have the usual cost structure (GPU servers) so there's nothing to subsidise by paywalling. Our paid tiers are for larger files, batch mode, and API access — not for gating the tools you already use.

Why does it take ~10 seconds the first time?

The AI model (~30 MB of compressed neural-network weights) downloads to your browser the first time you visit the tool. Your browser caches it. Every visit after that starts instantly. It's literally faster after that than the cloud competitors because there's no upload round trip.

Can I use the output commercially?

Yes. You own the image you dropped. The AI didn't contribute creative content; it just segmented what was already there. Same rule you'd apply to Photoshop's "Subject Select" feature.

Does this leak my image anywhere?

No. Open your browser's network inspector while you run a conversion — you'll see zero requests after the model loads. The only thing we know is that someone visited the page. We don't know what you dropped, what came out, or whether you downloaded it.

What about video? Can I remove backgrounds from clips?

Not yet. Segmenting 30 frames per second in a browser is still rough in 2026; we'll add it when the model performance crosses the threshold where it's actually usable. Track the roadmap if you're waiting for this specifically.

Ready?

Remove background →. No sign-up. No upload. Your image never leaves your machine.