Image Resizer

Four ways to make a picture fit a box

"Fit inside the box" scales the picture down until both sides are inside, so nothing is lost but one side may be shorter than the box - this is what a blog image wants. "Fill the box" scales until the picture covers the box and then crops the parts that stick out, so the output is exactly the box and nothing is distorted - this is what an avatar or a product thumbnail wants. "Stretch to the box" hits the box exactly by changing the shape of the picture, which is right for a banner and wrong for a face. "Percentage" ignores the box and scales by a number, which is what "make everything half size" literally means.

The output is never distorted unless you ask for it: a 1600 x 1200 photo fitted into an 800 x 800 box becomes 800 x 600, not a squashed 800 x 800. The plan on the page says which size you will get before you run anything, and the result table reports the size that was written.

A shape preset - square, 4:3, 16:9, 4:5, 9:16 and the rest - fixes the height from the width, which is how cropping to a social media shape actually gets asked for. 1080 px wide with a 1:1 shape is a 1080 x 1080 square, and the plan prints that.

Where the pixels go

Shrinking averages: every output pixel is the weighted average of the block of source pixels it covers, so a 1600-pixel photo cut to 800 keeps the information from all 1600 columns instead of throwing every second one away. Throwing them away is what makes naively resized pictures look jagged. Enlarging interpolates between the pixels that are there, and the page says plainly that this cannot create detail the camera never recorded.

Rotation and mirroring happen before the resize, and the Exif orientation flag that phones write is turned into real pixels first. That matters: a portrait photo from a phone is often stored landscape with a "turn this" note attached, and resizing before applying that note produces a sideways file that other programs then turn back, usually the wrong way.

Cropping is a rectangle taken out of the source, positioned by the anchor you choose - center, an edge or a corner. Cropping throws pixels away for good, which is why the page says so rather than presenting it as a pure win.

What changes beyond the pixels

Resampling rewrites every pixel, so the file is re-encoded and everything that was not pixels is dropped: GPS coordinates, camera details, timestamps, the color profile. Rotation is baked in, so the picture is upright everywhere. If the output format is JPEG, see-through pixels get a white background, because JPEG has no alpha channel at all.

The file size can go up even when the picture gets smaller - converting a JPEG to PNG, or enlarging, both do that - and the results table shows the real numbers instead of promising a saving. When the aim is a smaller file rather than a specific pixel size, the image compressor is the right page; this one is about dimensions.

The limits, and when the page says no

A job takes up to twenty pictures, each up to the file size on the upload box, with no more than 40 megapixels or 20000 pixels a side in the output. Those are practical ceilings - the working pixels sit in memory - and a picture that breaks one is named with the number it exceeded rather than quietly skipped.

The same honesty runs the other way: nothing is upscaled, because interpolation cannot create detail the camera never recorded. A resizer can only move the pixels that exist.