Image Resizer — Scale, cap or fix dimensions right in your browser

Resizing changes how many pixels an image has, not what is in it. Scale everything by a percentage, or limit just one side: set a max width and the height follows, set a max height and the width follows. Cap the long edge to normalise a set of assets, or the short edge for wallpapers and posters. For an exact size switch to fixed dimensions and pick how the image fills the box — stretch it and it distorts, fit it and the empty space takes a white, transparent or custom colour, cover it and the overflow is cropped away. Presets cover avatars, product shots, article and video covers plus 4K, 2K and 1080p, and fill both fields in one click. A no-upscale switch keeps small originals from being blown up and smeared. Compare side by side or drag a split line, magnify to check for jaggies, and switch on the checkerboard to inspect transparency. Download one file, zip them all, copy the image or its Base64 string; output names carry the new size by default so you can see at a glance what each one became, batch mode lets every file take its own size, and a CSV manifest keeps the books straight. Everything runs locally in your browser — nothing is uploaded, it works offline, and it adapts to phone, tablet and desktop.

Import images

Drop images here, or click to chooseYou can also press Ctrl+V to paste a screenshotJPG / PNG / WebP / AVIF, depending on your browser · Up to 40.00 MB per imageChoose images

Size mode

Common ratios

Scale50%

Size presets

Avatar

Product

Article cover

Short-video cover

One click applies a preset — it switches to “Fixed size” and keeps the current fill mode

Details

Never upscale

Leaves small originals at their own size instead of blowing them up

DPI tag

Writes a print-density tag into the file; on-screen display is unaffected

Output format

Keeping the original is simplest; choose PNG or WebP for transparency, JPEG has no alpha

Encode quality92%

Affects file size and detail only, never the dimensions

File name

Prefix

Suffix

{w} and {h} expand to the output width and height

Example: photo_1080x1080.jpg

How to use

Drop your images into the box above, click “Choose images” to pick several at once, or press Ctrl+V to paste a screenshot. To fetch from the web, paste the address into the URL field. Phone photos are auto-rotated from EXIF, so no manual rotating is needed.
Pick a size mode. “Percentage” scales everything and is the quickest route — 25%, 50% or 75% is one click. “Max width” or “Max height” limits a single side and lets the other follow. Use “Long edge” to normalise a set of assets, or “Short edge” for wallpapers and posters.
For an exact size choose “Fixed size” and then decide how to fill the box: “Stretch” covers it but distorts, “Fit” keeps the aspect ratio and pads the rest with your background colour, “Cover” fills the box and crops the overflow. If that feels like work, tap a preset below — avatars, product shots and covers are all ready to go.
Compare the original and the result on the right; switch to “Slider” and drag the divider to inspect the same spot in both, or press and hold to magnify and check for jaggies. When it looks right, download it, zip several files, or export a CSV manifest. File names carry the output size automatically.

Frequently asked questions

What is the difference between Max width and Fixed size?

Max width pins only one side — the height is computed from the original aspect ratio, so the image never distorts. Fixed size pins both values, and when the original ratio does not match you have to choose how to fill the box: Stretch distorts, Fit adds padding, Cover crops the overflow. Use Fixed size when a set of assets must all be identical; use Max width when you only need to keep an image from being too large.

I typed 800×800 but the export is not 800×800. Why?

Most likely Never upscale is on. When the original is smaller than the target, it would rather output at the original size than stretch the image, because stretching only produces mush. Turn Never upscale off and it will follow your numbers exactly — at a visible cost in sharpness.

Can I get rid of the white bars that Fit leaves behind?

Those bars are the price of preserving the aspect ratio — as long as nothing is cropped, they will be there. If you need the frame filled with no bars, switch to Cover, which crops the overflow to the target ratio. Stretch also leaves no bars, but people and objects get squashed, so it is rarely the right answer.

Does resizing make an image blurry?

Shrinking is lossy, but at modest ratios you will not see it — the more you shrink the more detail goes, though anything under about half is usually indistinguishable by eye. Enlarging is always blurry, because those pixels never existed and can only be guessed by interpolation. Shrink rather than enlarge, and start from a larger original when you must.

In a batch, can each image have its own size?

Yes. Every image shares one rule by default, but clicking a row lets you type a width and height just for that file; leave one side blank and it still follows the ratio. Individually sized files are marked, and Reset to shared rule puts them back without touching the others.

What is it

Scaling is resampling: work out which spot in the original each output pixel corresponds to, then interpolate a colour from the pixels around it. Shrinking squeezes several source pixels into one, so information is lost and edges turn jagged or moiré; enlarging spreads one pixel over many, so the in-between colours have to be guessed — guess smoothly and it blurs, guess sharply and it blocks up. Resizing is therefore always lossy, only to different degrees, which is why blowing a small image up and shrinking it back never returns the original quality.

Origin & history

The earliest resampling method was nearest neighbour: each output pixel simply copies the closest source pixel. That is what the digital image-processing texts of the 1970s described, because it cost almost no time on the machines of the day. The flaw is obvious — shrinking throws away whole patches of detail, enlarging turns every pixel into a square. In the late 1970s and early 1980s bilinear and bicubic interpolation arrived: the first averages the surrounding 2×2 samples, the second uses a 4×4 neighbourhood with a cubic convolution kernel and is visibly sharper. In 1981 Robert Keys at SGI published the unified derivation of cubic convolution, and the kernel in that paper is still the mathematical basis of Photoshop's bicubic resampling. That same year Lanczos resampling was applied to images; it windows with a sinc function for the highest acutance, at the price of ringing, which shows up as faint halos around fine edges in thumbnails. The real answer to “why does shrinking produce moiré” came in 1983, when Lance Williams proposed the mipmap: rather than sampling arbitrarily when shrinking, precompute the image as a chain of half-size levels and pick the appropriate one. That structure ended up in every GPU and explains why textures in games do not shimmer at a distance. Seam carving brought content awareness in 2007, using an energy function to decide which pixels may be deleted, and became the standard example of non-uniform resizing. In 2014 SRCNN applied a three-layer convolution to super-resolution, and ESRGAN added adversarial training in 2018 — the first time scaling started inventing detail rather than interpolating it. Canvas scaling in browsers today still uses only bilinear and bicubic: fast, stable, predictable. Which is exactly why forcing a small image larger always looks soft. A footnote worth knowing: the imageOrientation option on createImageBitmap lets EXIF rotation be corrected at decode time, which finally closed the long-running sideways-photo problem.