Why is my JPEG so big?

The short answer

There are five reasons a JPEG comes out larger than you expected. They are not equally important, and the one people reach for first is fourth on the list.

1. Too many pixels. By far the biggest, and the easiest to fix.
2. Noise or grain in the image, which is invisible and enormously expensive.
3. A quality setting higher than anything can perceive.
4. Embedded previews and metadata riding along inside the file.
5. It should not be a JPEG at all.

Everything below is measured rather than asserted. The test image is a 4000 × 3000 photo-like scene, encoded in a browser, and the method is described where it matters.

1. Too many pixels

A recent phone camera produces roughly 12 megapixels. A photo on a website is usually displayed somewhere between 800 and 1600 pixels wide. You are storing several times more detail than will ever reach anyone.

Here is the same photograph scaled down and saved at quality 85 each time.

Halving the width and height took the file from 2,528 KB to 352 KB. That is 86 percent gone for an image that still looks identical at any normal viewing size.

Note that it beat the obvious prediction. Halving both dimensions leaves a quarter of the pixels, so you might expect a quarter of the file, around 630 KB. It came in far below that, and the reason matters for the next section: scaling an image down averages neighbouring pixels together, which quietly removes the fine random detail that costs the most to store. Resizing does two useful things at once.

2. Noise you cannot see

This is the cause almost nobody mentions, and on photos taken indoors or at night it is often the dominant one.

Compression works by predicting detail from surrounding detail and storing only the difference. Noise is random, so it cannot be predicted, so every speck of it has to be stored individually. Grain is the most expensive thing you can put in a photograph.

Same scene, same 1600 × 1200 dimensions, same quality 85. The only difference is how much random variation was added to each pixel, on the 0–255 scale.

A variation of ±13 out of 255 is about five percent. On screen it reads as very slight texture. It cost five times the file size.

This is why the photo you took in a dim restaurant is four times bigger than the one you took outside at noon, despite identical camera settings and dimensions. The camera amplified its sensor to cope with the darkness, and amplification produces grain.

There is no clean fix inside a JPEG, which is precisely why resizing helps so much here: scaling down averages adjacent pixels, and averaging random noise is how you remove it.

3. The quality setting is higher than anyone can see

Plenty of software defaults to quality 95 or 100 when exporting. In separate testing, quality 100 produced a file three times the size of quality 90 for an average pixel difference of about one percent.

Quality 100 is not lossless. It is the same lossy process applied so gently that the format stops being efficient, while still discarding data. For anything heading to a screen, 80 to 85 is where almost all the size disappears and almost none of the visible quality does.

The full measurements are in does compressing an image reduce quality, along with what happens when you re-save the same file fifty times.

4. Things riding along inside the file

A JPEG from a camera is not only image data. It carries EXIF metadata, and it frequently carries a second copy of the picture at reduced size, which exists so that photo software can show you a preview without decoding the whole thing.

EXIF text is genuinely small, a few kilobytes at most. The embedded preview is not. Here is what previews at typical sizes cost:

Whether that matters depends entirely on what it is attached to. Ninety-five kilobytes inside a 2,528 KB original is four percent and not worth thinking about. The same 95 KB attached to a photo you have already resized to 186 KB is a third of the file. Metadata stripping is a finishing step, and it is worth doing last rather than first.

5. It should not be a JPEG

JPEG is built for photographs. Applied to screenshots, diagrams, logos or anything with text, it does two bad things at once: it smears the edges of the text, and it does not even save you much space.

We ran a screenshot-like image, full of flat colour, ruled lines and small text, through both formats at 1600 × 1200:

For the photograph, JPEG is ten times smaller than PNG and the choice is obvious. For the screenshot the two are within a few percent of each other, and once you ask for quality 95 the JPEG is actually the larger file. So JPEG buys you nothing there, and you pay for it in smeared text.

The rule that follows is simple: photographs are JPEG, everything else is PNG. If your oversized JPEG is a screenshot, converting it with JPG to PNG will make it look better, though be aware it will not undo the smearing already baked in by the first save.

Fixing it, in order

The order matters more than any individual step, because each one changes how much the next is worth doing.

1. Check the dimensions. Anything over about 2000 pixels wide that is destined for a screen is carrying weight it does not need. This is the biggest single win and the only one that costs no quality.
2. Resize to what will actually be displayed. 1600 pixels wide is generous for a full-width web image, 800 is plenty for something in a column.
3. Then set quality to 80 to 85. Not before resizing, or you will be judging quality on an image that is about to change.
4. Then strip metadata if you still need the last few percent, or if the photo was taken somewhere you would rather not publish.
5. If it is not a photograph, stop and use PNG instead.

That order is not arbitrary. We tested the two big steps against each other by hitting the same file size each way, and the results are in is it better to resize or compress an image.

Frequently asked questions

Why are my JPEG files so big?

Almost always because the image has more pixels than it needs. A modern phone photo is around 12 megapixels, and file size scales with pixel count, so a picture destined for a website or an email is typically carrying five to ten times the detail it will ever display. The second most common cause is sensor noise from low light, which is nearly invisible but extremely expensive to store.

Does removing metadata make a JPEG smaller?

A little, and it matters more than you would think on an already-small file. Camera EXIF data itself is only a few kilobytes, but many cameras and phones also embed a full preview image inside the file. We measured what those previews cost: around 4 KB for a small thumbnail, but roughly 95 KB for a 1024 pixel preview. On a 2.5 MB original that is noise. On a photo you have already resized down to 190 KB, it is a third of the file.

What should I do if a JPG file is too big to email or upload?

Resize before you compress. Check the pixel dimensions first: if the image is wider than about 2000 pixels and it is going on a screen, reducing the dimensions will save far more than any quality setting, and it does it without introducing compression artifacts. In our testing, halving a photo's width and height cut the file by 86 percent. Only after resizing is it worth reaching for the quality slider.

Why is my photo huge when it was taken in a dark room?

Low light forces the camera to amplify its sensor signal, and that amplification produces grain. Fine random detail is the single most expensive thing a JPEG can store, because compression works by discarding predictable detail and noise is by definition unpredictable. We measured the same image at the same size and quality with and without grain: 81 KB clean, 425 KB with moderate noise added. More than five times the size for something you can barely see.

How do I reduce a JPEG to a specific file size?

Work in that order: reduce the dimensions to what will actually be displayed, then lower the quality setting until you hit your target, then strip metadata if you still need the last few percent. Doing it the other way round means crushing the quality of an image that was simply too large to begin with, which is what produces the blocky results people associate with compression.