Skip to content
Server-sideDeleted in 30 minutes

Every Channel Wants a Different Shape, and Cropping Loses Products

A tall bottle photographed in portrait and a wide keyboard photographed in landscape cannot both be center-cropped into a square without one of them losing something. The crop takes the same proportion from every image regardless of what is in it, so the bottle keeps its label and loses its cap while the keyboard loses both ends. Fitting inside the box instead — scaling until the longest side reaches the limit and leaving space around the rest — costs nothing and cuts nothing off.

Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.

Add to Chrome — free

What it does

  • Resize by width, height, percentage, or fit box
  • Many files at once, with per-file results
  • No account, no watermark, deleted within 30 minutes
  • Aspect ratio preserved unless you change the mode

How to use Resize Product Images

  1. 1

    Collect the target sizes first

    Each channel publishes its own dimensions. Write them down before you start, because resizing to the wrong number twice takes longer than reading the documentation once.

  2. 2

    Fit inside the box rather than filling it

    Choose the mode that scales the image to fit the target dimensions. That keeps the whole product in frame and leaves the padding to the platform or to a separate composite.

  3. 3

    Check the smallest output at its real size

    A thumbnail is the size most customers see. View it at that size before accepting the batch, because detail that disappears there was never load-bearing.

How it works

This page runs the same resizer used elsewhere in the catalog, set up for product work. Files are decoded on your device, drawn into a canvas at the requested size, and encoded back out — and a batch is processed one file at a time locally.

Reduction is done in steps rather than in a single jump, because a large one-shot downscale skips over detail that a stepped one averages properly. The difference shows up as aliasing on fine patterns, which product photography has a lot of.

The output is a zip when you process several files, with each result named after its source.

Fit, fill, and the difference that matters

Three ways of getting an image to a target size, and only one of them is safe to apply blindly.

Fit scales until the whole image sits inside the target box, so the longest side hits the limit and the other is shorter. Nothing is lost, and the result is not the exact target dimensions unless the proportions already matched.

Fill scales until the box is covered and crops whatever hangs over. The result is exactly the target dimensions and part of the image is gone. Applied to a set, what gets cut depends entirely on each image's proportions.

Stretch forces both dimensions and distorts. Nobody wants this, and it happens regularly by accident.

For products, fit is the default because the channel usually pads for you and because a bottle with its cap cut off is worse than a bottle with white space beside it. Fill is a per-image decision, not a batch one.

The sizes are published, and they change

There is no universal product image size, and guessing costs a re-run.

Marketplaces state a minimum for zoom and a recommended dimension. Store platforms state a maximum they will serve and generate their own thumbnails below it. Ad platforms state exact dimensions per placement and reject anything else. Social channels change theirs more often than anyone would like.

The practical approach is to keep one master file per product at the largest sensible size, and generate channel sizes from it as needed. Resizing down from a master is lossless in the sense that matters — you can always do it again — while a channel-sized file that has been resized twice cannot be recovered.

Enlarging is not upscaling

Worth separating, because the word is used for both.

Making an image larger in this tool redistributes existing information across more pixels. Nothing is added, and the result is exactly as soft as the original was, just bigger. That is the correct behavior for a resizer and it is what you want when a channel demands a minimum dimension you narrowly miss.

Adding detail that was never captured is a different operation, done by a model that has learned what detail looks like, and it belongs on the upscaling pages. For products it comes with the caveat those pages carry: invented detail on a product is invented information about a product.

If a listing image is genuinely too small, the fix that works is rephotographing it.

What has been tested

The resizer this page uses is already in production, and it was exercised with real files before this page went up.

A mixed pair — one landscape 2400×1600 and one portrait 1200×1800 — was processed in one batch, producing per-file results and a zip. Fit, fill, percentage, and exact-size modes are all present, so the distinction this page draws between them is a real choice in the interface rather than a description of resizers in general.

Enlargement was confirmed separately: 1200×1800 to 2000×3000, with "Never enlarge" unticked, and the result exactly as soft as the source.

Still to check on a later pass: a batch containing a transparent PNG, and the smallest output inspected at the size a customer's thumbnail actually renders.

Examples

A range in mixed orientations

products/*.jpg - portrait and landscape, various sizes
resized-images.zip - each fitted inside the same box

The intended case. Fitting rather than cropping means the tall and the wide products both arrive complete, at a common maximum dimension.

One image up to a larger size

small.jpg - 600x600
small-2000.png - 2000x2000

Possible and rarely wise. Enlarging redistributes the pixels you have; it does not add detail, so the zoom view stays as soft as the source.

Frequently asked questions

Should I crop to square instead?

Only with a per-product decision. A center crop applies the same rule to every image and does not know where the product is, so it works for a range of similar objects and quietly ruins anything long or unusually shaped. Crop deliberately, in the crop tool, or fit and pad.

Does resizing lose quality?

Reducing loses detail by definition, and doing it in one large step loses more than stepping down gradually, which is why this tool steps. Enlarging cannot add detail at all — it makes the same information occupy more pixels, which reads as softness.

What format should product images be?

JPEG for photographs, PNG when you need a hard-edged cut-out or transparency, WebP if every channel you use accepts it. Photographs in PNG are usually several times larger for no visible gain, which on a collection page is a real cost.

Is anything uploaded?

Yes. Each photograph is sent to our server, resampled with libvips, and returned. The upload is deleted as soon as the resize finishes and the result on a thirty-minute timer, so nothing survives past about thirty-five minutes, and there is no account or history. The reason for doing it here rather than in the browser is quality: libvips applies a proper Lanczos filter, where a browser canvas does bilinear scaling that visibly softens any reduction of more than about half.