Image guide · Updated September 19, 2026
Resize or compress: which change do you need?
Pixel dimensions describe the image grid. File size describes the bytes used to store it. They solve different problems.
Compression keeps the grid unless you ask otherwise
With Original dimensions selected, a 1200 × 800 input keeps that pixel grid. A lossy JPEG, WebP, or AVIF output can use fewer bytes by representing detail less precisely. Inspect edges, skin, and gradients in Compare. PNG optimization instead tries to represent the same prepared pixels more efficiently.
Resizing removes pixels
Changing a 1200 × 800 image to 50% produces 600 × 400: half the width and height, but one quarter of the pixels. This does not promise a fourfold reduction in bytes; format, detail, and metadata also matter. Once detail has been removed, enlarging the smaller file cannot restore it.
Fit within a box without stretching
Choose Fit within dimensions and keep the aspect-ratio lock on. A landscape image fits inside both maxima; it does not necessarily fill the box. For example, 1200 × 800 fitted inside 500 × 500 becomes 500 × 333 after rounding. The app does not crop it into a square or enlarge a smaller source to fill the box.
A practical order for web images
First choose dimensions that suit where the image will be displayed. Then choose the destination’s supported format. Finally reduce quality or set a file-size limit and inspect the result. For a high-density display, a 400 CSS-pixel placement may benefit from an 800-pixel source. That is a display choice, not a rule every upload needs.
When preserving dimensions matters
An ID form, print workflow, or marketplace can require minimum dimensions even when it also imposes a byte limit. Start with compression alone in those cases. If the target fails, do not resize below the receiving service’s stated minimum. Our examples compare identical source pixels with explicit scale changes so you can see what each control did.
Try it with a reproducible example
Open the source files, measured outputs, and settings. Use the same source for each comparison and inspect the actual result.