When file size limits and dimension requirements fight each other

Most marketplace image requirements are easy on their own. The difficulty comes from combinations, and one combination in particular.

Walmart's own guidance recommends 2000 x 2000 pixels for zoom while capping file size at 1MB. Those two numbers pull in opposite directions, and on a detailed product they pull hard.

Why it is tight

A 2000 x 2000 photograph is four million pixels. At 1MB, that is about two bits per pixel, including all colour information.

For a simple product on a plain white sweep, that is comfortable, large flat areas compress to almost nothing, so most of the budget goes on the product itself. Quality 90 or higher will often fit.

For a detailed product, knitwear, printed packaging, anything with fine texture, it is not comfortable at all. Detail is precisely what compresses badly, and you can find yourself needing quality 60 or below to fit.

That matters beyond softness. Below about quality 60 with chroma subsampling on, and with the product near the frame edge, compression can push corner pixels off pure white, which then fails the background check on the same platform that imposed the size limit.

So the two requirements can genuinely conflict, and satisfying one naively breaks the other.

The levers, in the order to pull them

1. Give the product a margin. Moving the subject away from the frame edge removes the chroma-artefact problem outright, and costs nothing. If the platform asks for 85% fill, you have room.

2. Turn off chroma subsampling. Counter-intuitive, because it makes the file bigger. But it eliminates the coloured-edge artefacts that cause corner failures, and on a mostly-white image the cost is smaller than you would expect, often only 5 to 10%.

3. Lower quality in steps, and record where you landed. Walking down from 92 in steps of 4 or 5 until the file fits is the right approach. A tool that does this silently is fine; one that tells you it settled on 58 is better, because that is the number that predicts a corner failure.

4. Simplify the background before compressing. A perfectly uniform white background compresses to almost nothing. A background with a gradient, sensor noise or a faint shadow does not. Compositing onto flat white before encoding often buys more than two quality steps, and it is free, because you wanted a white background anyway.

5. Reduce dimensions, last. Dropping to 1600 x 1600 will fit easily, and it costs you the zoom feature. Only worth it if the alternative is genuinely artefacted.

Why the order matters

Steps 1 and 4 cost nothing at all. Step 2 costs a little size but buys correctness. Step 3 costs visible detail. Step 5 costs a feature shoppers use.

Most people go straight to step 3, drop quality until it fits, and never consider that flattening the background would have saved them fifteen points of quality for free.

Numbers from a real jumper

A knitted jumper, photographed on white, at 2000 x 2000.

• Straight export at quality 92: about 2.4MB. Too large.
• Quality 70: about 1.3MB. Still too large.
• Quality 58: about 0.95MB. Fits, but the weave is visibly mushy and one corner reads 252.
• Background composited to flat white first, then quality 82: about 0.9MB. Fits, weave intact, corners at 255.

The difference is not the compression settings. It is that the first three attempts were spending most of their bit budget encoding a background that did not need encoding, sensor noise across a large area is expensive, and it was noise nobody wanted.

Where the budget actually goes

It helps to know roughly how a JPEG spends its bytes on a white-background product shot.

The background, if it is genuinely flat, costs almost nothing, large uniform areas are what the format handles best. The product edge costs a surprising amount, because a hard transition is high-frequency information and high frequencies are what compression discards last if you want them to survive. Fine texture across the product costs the most of all.

Sensor noise is the hidden expense. It looks like nothing and behaves like texture, so it consumes bits everywhere it appears, including across a background where nobody wanted any detail at all. On a shot taken at ISO 1600 rather than ISO 100, noise alone can add 20 to 30% to the file for no visible benefit.

Which gives one more lever, upstream of all the others: shoot at base ISO. It is free, it costs nothing in the export, and it makes every subsequent compression decision easier.

The general principle

Spend your bits on the product.

Anything in the frame that is not the product is competing for the same budget: noise, shadow gradients, texture in the paper, a faint vignette. Removing them is not a compression trick, it is removing information the buyer was never going to look at.

Which is a reasonable argument for doing background replacement even where the platform does not require pure white. A flat background is a smaller file at the same quality, and a smaller file at the same quality is a sharper product at the same limit.