Skip to content
Pixora

Remove Image Background

Processed locally in your browser — nothing is uploaded.

Quick answer

Cut the subject out of a photo and get a transparent PNG, or drop it straight onto a solid color. The AI model runs inside your browser tab, so the photo never leaves your device and there is no account, no credit system and no per-image charge.

What the remove image background does

Removing a background is not a filter, and there is no clever pixel trick that does it. Deciding which pixels belong to the subject and which belong to what is behind it is a perceptual judgement, and the only tools that make it well are neural networks trained on hundreds of thousands of images. This page runs U-2-Net, an open-source salient object detection model published in 2020 and still one of the most widely used in the field. It looks at the whole image at several scales at once, which is how it can tell that a dark jacket against a dark doorway is still a jacket.

The important difference here is where the model runs. Every mainstream background remover sends your photo to a server, does the work on their hardware, and sends a cut-out back. That is why they need an account, why they meter you, and why they can charge per image. It also means a copy of your photo sat on someone else’s disk, at least briefly, and you have only their word about what happened to it next.

Pixora ships the model to the photo instead. The weights are about 4.5 MB and are downloaded once, only when you open this page, then cached by your browser. Inference happens in a Web Worker on your own CPU. Open your developer tools, watch the Network tab, and remove a background: you will see the model come down, and you will not see your image go anywhere. Nothing is uploaded because there is nowhere for it to be uploaded to.

Be realistic about quality. A 4.5 MB model that has to run on a phone is not going to match a service burning a datacenter GPU on every request. On people, products, pets and anything with a reasonably clear edge, the result is clean and usable straight away. On flyaway hair, fur, motion blur, glass, smoke and semi-transparent fabric, it will be visibly rougher than the paid tools. The edge threshold slider exists precisely for those cases, and it is worth two or three attempts before you conclude the tool cannot do it.

How to use it

  1. Add your photo

    Drop the image onto the page or click to browse. The AI model downloads on first use and is cached after that.

  2. Wait a few seconds

    The model runs on your own device, so it takes a few seconds per image rather than being instant. A phone is slower than a laptop.

  3. Tune the edge if you need to

    Raise the threshold if background is bleeding into the subject; lower it if parts of the subject are being cut away.

  4. Pick transparent or a solid color

    Leave it transparent for a PNG you can composite anywhere, or choose a background color for a ready-made product shot or profile picture.

  5. Download

    Transparent results must be PNG or WebP. JPEG has no alpha channel, so a transparent JPEG is not a thing that exists.

Your images never leave your device

This is the tool where the privacy difference stops being abstract. Background removal is what people reach for with passport photos, ID shots, product photography before a launch, headshots, and pictures of their children. Those are exactly the images you would least like to hand to a third party, and they are exactly the images the mainstream removers require you to upload.

Because the model executes on your own hardware, none of that leaves the tab. There is no account to create, so there is nothing to link an image to a person; there is no server-side queue, so there is no copy sitting on a disk awaiting deletion; and there is no per-image credit, so nothing has to be logged in order to bill you for it. The absence of a backend is not a limitation we are apologising for. It is the feature.

  • No file is ever uploaded to a server
  • Works offline after the first visit
  • No account, no watermark, no limits

Format and quality tips

The threshold slider is not a quality dial

It sets where the model’s confidence stops counting as background and starts counting as subject. Raise it when a busy or high-contrast background bleeds into the cut-out; lower it when the edges of the subject, or thin things like a strap or an antenna, are being eaten. There is no globally correct value, which is why it is exposed rather than hidden.

A plain, contrasting background is worth more than any setting

If you are the one taking the photo, you have far more leverage than the model does. A subject shot against a plain wall in even light will cut out almost perfectly; the same subject against a cluttered room at dusk will not, no matter what you drag. Photographers have known this since long before anyone trained a network, and it is why green screens exist.

Never save a cut-out as JPEG

JPEG has no alpha channel. Export a transparent image as JPEG and the transparency does not fail loudly, it fills in, usually with white or black, and you are left with the exact rectangular background you just spent time removing. Use PNG, or WebP if file size matters.

Hair is the honest limit

Fine hair, fur and anything semi-transparent are where a compact model shows its size. If the result is not good enough for the job, that is a real answer and not one we will hide from you; a photo where the edge matters that much probably wants a manual mask in a full editor.

Frequently asked questions

Is my photo uploaded to remove the background?

No. The AI model is downloaded to your browser and runs on your own device, which is the opposite of how every mainstream background remover works. You can confirm it yourself: open developer tools, watch the Network tab, and remove a background. You will see the model download and you will not see your image leave.

Why does the first image take longer than the rest?

The first one has to download the model, roughly 4.5 MB, and compile it. After that it is cached by your browser and every subsequent image skips straight to the work, so the second image is much faster than the first.

Is this as good as remove.bg?

Honestly, not quite. Those services run far larger models on datacenter GPUs. On people, products and anything with a clear edge you will struggle to see a difference, but on fine hair, fur and semi-transparent material a paid service will produce a cleaner edge. The trade you are getting in return is that your photo is never uploaded, there is no account, and there is no per-image charge.

Why is my transparent background white when I open the file?

Almost always because it was saved as JPEG, which has no alpha channel and fills transparency with a solid color. Save as PNG or WebP instead. Some viewers also paint a white sheet behind transparent images, so it is worth checking the file in a browser before assuming it is broken.

Can I remove backgrounds from a batch of photos?

Yes. Add as many as you like and they are processed one after another on your device, then downloaded as a ZIP. Bear in mind each one takes a few seconds of real computation, so a large batch takes real time, unlike a compression job.

It cut off part of my subject. How do I fix it?

Lower the edge threshold, which makes the model more willing to keep pixels it is less certain about. If a whole limb or object is missing, the model has not recognised it as part of the subject at all; cropping tighter around the subject before running the tool often fixes that.

Further reading

Related tools