PNG supports transparency. Converting to PNG does not create any
This is the sentence that would save more frustration than anything else on this page, so it goes first.
Most people who convert a JPG to a PNG are trying to get rid of a background — usually a white one behind a logo, a signature or a product shot. They convert, open the result, and it looks exactly the same. Nothing has gone see-through. The file is simply larger.
That is not a broken tool. A JPG has no transparency anywhere in it, because the JPEG standard has no channel to store it in. PNG can store transparency — that is the whole reason to move to it — but changing the container does not decide which pixels should become see-through. Something has to make that decision, and a plain conversion does not.
So this page does both jobs. It converts, and it will knock the background out for you if you ask it to. The switch is on the tool above, and the picture updates as you move the sliders so you can see exactly what is going.
How knocking out a background by colour actually works
The method is simple to describe: pick the colour of the background, and every pixel close enough to it becomes transparent. Two words in that sentence are doing all the work — close enough.
A JPEG background is never one exact colour. JPEG compresses in eight-by-eight blocks and throws away fine detail to do it, so a flat white comes back as thousands of slightly different near-whites. A photographed white has more against it still: a vignette from the lens, noise from the sensor, and a slight colour cast from whatever was lighting it.
Measured on a realistic example while building this page — a white background with a gentle vignette and ordinary sensor noise, saved at normal JPEG quality:
Softness handles the other half of the problem. The pixels right at the edge of a subject are a blend of the subject and the background — that is what makes edges look smooth rather than jagged. Cut them off sharply and the result has a hard, cut-out-with-scissors look. Instead of one threshold, this tool uses two: everything within the tolerance goes completely, everything beyond tolerance-plus-softness stays completely, and the band between them fades. That band is what keeps a knocked-out logo looking like it belongs wherever you place it.
The colour matching itself does not use plain arithmetic on the red, green and blue numbers, because the eye does not weigh those three equally — a shift in green is far more visible than the same shift in blue. It uses a weighted distance of the kind Thiadmer Riemersma describes, which is cheap to compute and much closer to how a difference actually looks. In practice that means a tolerance high enough to clear a blue background will not also eat a green subject.
When this works beautifully, and when it will not work at all
Colour keying is an old, reliable technique with an honest limitation: it can only separate things that differ in colour.
It works well on a logo on a solid background, a signature on white paper, a scanned document, a product photographed against a studio backdrop, a screenshot with a uniform surround, and clip art.
It will not work on a person photographed against a busy street, a white dog on a white sofa, or anything where the subject contains the same colour as the background — because the tool cannot tell which white you meant. It will also struggle with soft shadows, motion blur and fine strands of hair, where thousands of pixels are genuinely part background and part subject.
For those, the right tool is one that understands what a subject is rather than what colour it happens to be — the AI Background Remover on this site does that job. The distinction is worth knowing: colour keying is instant, precise and predictable within its limits; subject detection copes with complexity but is slower and occasionally wrong in ways you have to check.
Why the PNG comes out so much bigger, and what to do about it
Converting a JPG to a PNG reliably produces a much larger file. Measured here: a 12,328-byte JPEG became a 66,177-byte PNG — five times the size, looking identical.
Nothing has gone wrong. JPEG got small by discarding detail. PNG is lossless, as its specification requires, so it faithfully stores everything it is given — including all the compression artefacts the JPEG left behind. You are paying full price to store an already-degraded picture perfectly.
That is the cost of transparency, and it is usually worth paying. But there is a cheaper way to pay it: lossless WebP holds exactly the same pixels, transparency included, and is typically far smaller than the equivalent PNG. Every current browser reads it, as caniuse records. The tool above offers it as an alternative output for exactly this reason. Choose PNG when the file is going somewhere you cannot test; choose lossless WebP when it is going on a website.
Questions people ask about JPG to PNG
Will converting to PNG improve my image quality?
No, and nothing else will either. Whatever the JPEG discarded is gone. PNG will preserve what remains exactly, which is useful if you are about to edit and re-save several times — each of those saves will then cost nothing further — but it cannot restore anything.
How do I make the background transparent?
Turn on the switch in the tool above, check the colour it picked from the corners, and adjust the tolerance until the chequerboard shows through cleanly. You can also click anywhere on the preview to take the colour from that exact spot, which is the quickest way when the background is not what the corners suggest.
Why did my whole picture disappear?
The tolerance is too high, or the colour picked is one your subject shares with the background. The tool tells you what percentage went see-through and warns when almost all of it did — but the preview is the real check, so watch it as you move the slider.
Why is some of the background still there?
The tolerance is too low for how much that background varies, or there is more than one background colour — a gradient, or a shadow. Raise the tolerance a little at a time. If the subject starts to erode before the background clears, the two are too similar for colour keying and you want the AI remover instead.
Can I convert several files at once?
Yes, up to 30, all with the same settings. The preview and the knock-out settings are worked out from the first picture, so this is most useful for a batch shot against the same background — a set of product photographs, for instance.
Does converting remove the EXIF data?
Yes. The picture is redrawn to make the conversion, and what comes back is pixels alone — no camera model, no date, no GPS coordinates. That is a privacy gain if you are about to publish the picture and a loss if you were relying on it.
Is PNG better than JPG?
Neither is better; they answer different questions. PNG stores every pixel exactly and carries transparency, which suits logos, screenshots, diagrams and line art. JPEG discards detail cleverly, which suits photographs and produces far smaller files for them. Converting a photograph to PNG is usually the wrong direction — see the PNG to JPG tool for the other one.
What about transparency in JPG — can it be added another way?
No. JPEG has no alpha channel, full stop. Any tool that appears to give a JPG a transparent background has quietly saved it as something else. If you are told you have a “transparent JPG”, check the actual file extension.
Are my images uploaded?
No. Everything — the conversion and the background removal — happens inside this page on your own machine. Nothing is sent anywhere and nothing is stored.
Where the facts on this page come from
- PNG Specification (Third Edition), W3C, on lossless compression and the alpha channel.
- ITU-T T.81, the JPEG standard, which defines no transparency channel.
- Colour metric, Thiadmer Riemersma, on weighting red, green and blue differences the way the eye does — the distance measure the knock-out uses.
- MDN — pixel manipulation with canvas and getImageData(), which is how the transparency is applied here.
- MDN — image file types and formats, and caniuse — WebP for support.