Why photos appear sideways or upside down, and how to fix them for good

The camera doesn't always turn the picture

When you hold a phone sideways or upside down, the camera sensor is turned too. Rather than rotating the picture data, a camera can save the image as the sensor captured it and add a note saying which way up it should be shown. The Exif standard defines an Orientation tag for this: it records the orientation of the camera relative to the scene when the photo was taken.

The jpegclub.org page on Exif orientation explains that a camera can set this tag from its orientation sensor without transforming the image data itself. The photo is stored one way and displayed another.

The eight orientation values

The tag holds a number from 1 to 8. Microsoft's developer reference for photo orientation describes each one as an instruction for how to rotate the photo to display it correctly:

• 1: no rotation needed; the photo is already upright.
• 3: rotate 180 degrees.
• 6: rotate a quarter turn clockwise (Microsoft writes it as 270 degrees counter-clockwise).
• 8: rotate a quarter turn counter-clockwise.
• 2 and 4: flip horizontally or vertically, a mirror image.
• 5 and 7: a quarter turn combined with a horizontal flip.

Why some apps show it sideways

The tag only works if the app reads it. An app that ignores it shows the pixels exactly as they were stored, which for a photo taken with the phone held sideways means a picture lying on its side.

In web pages, following the tag is the default. MDN, the web documentation run by Mozilla, lists from-image as the initial value of the CSS image-orientation property: the Exif information in the image is used to rotate it appropriately. Not every program, upload system or image library does the same, and you can't tell from the outside which ones read the tag.

The problem can also be created by accident. A tool that removes all metadata, including the orientation tag, but leaves the stored pixels as they were, turns a correct photo into a sideways one.

The double-rotation trap

The opposite mistake also happens. If a program turns the pixels upright but leaves the old tag in place, an app that reads the tag turns the picture a second time, and it ends up sideways again or upside down.

The jpegclub.org page notes that the jpegtran tool deliberately does not change the tag during its lossless rotations, and that a separate script is used to rotate the image and then set the tag so it stays consistent. The goal of that script is a file whose Orientation becomes 1. Any correct fix ends the same way: pixels and tag must agree.

Two ways to fix it

Keep the tag. The pixels stay as stored and the tag keeps doing its job. Nothing is re-saved, so there is no quality cost, but you still depend on every app that opens the file reading the tag.

Save it upright. The picture is turned so the pixels themselves are the right way up, and the tag no longer matters. This works in every app, including ones that ignore Exif. For a JPG it means saving the picture again, which costs a little quality; the guide on image compression explains why.

For a photo going to an upload form, an older system or a printer, saving it upright is the safer choice.

What our tools do

Crop image, the converters, the compressor and the resizer all save a new picture. They apply the photo's orientation tag to the pixels, so the new file is saved upright and no longer needs the tag. In Crop image you can also use rotate left, rotate right and mirror if the photo was taken the wrong way up in the first place, and apply the same crop to many photos.

Remove EXIF data works the other way. It keeps the orientation tag on purpose while removing other details, so a photo that displays correctly before cleaning still displays correctly after. Everything runs in your browser and nothing is uploaded.