Image size as a variable

More pixels stop helping a rating tool once a photo clears the size the tool downscales to on upload. Below that threshold, a low-resolution image genuinely loses detail and the score can move; above it, extra pixels only make the file larger and the upload slower.

What gets lost at low resolution

A heavily downscaled photo loses fine texture and edge definition first, because those are the details that need the most pixels to survive compression and resizing. An axis that depends on reading that kind of detail is reading less of it in a low-resolution submission, and the resulting score reflects an image with less information in it - not necessarily a worse subject, a less legible photograph of the same one. This is a real effect, not a superstition: below some size, the image genuinely tells the tool less than the source photo could have. Degraded detail is a known weak point for vision models in general: Dodge and Karam (2016) found four deep networks susceptible to blur, noise and JPEG compression, particularly to blur and noise.

Why the effect disappears above a threshold

Every tool that accepts photo uploads resizes them on ingest to something the underlying model actually processes - models take a fixed input size, not whatever resolution arrived. What that model does with the frame it receives is a longer subject, but the direct consequence here is straightforward: once a submission is at or above whatever the tool downscales to internally, sending more pixels changes nothing the model sees. A 48-megapixel photo and a 12-megapixel photo of the same subject, both above that threshold, arrive at the model looking effectively identical. The only thing the extra pixels buy at that point is a slower upload.

The protocol implication

The instinct to "use the best camera available" is aimed at the wrong lever. Resolution is not something to maximise; it is something to clear a floor and then stop worrying about. Any recent phone camera clears that floor comfortably at its default settings, which means the resolution question resolves itself the moment a reasonable device is used and left on its normal settings. What actually needs attention is holding the setting constant afterward, the same as any other variable in a comparable submission - a photo taken at a much lower resolution than usual, for whatever reason, is not directly comparable to the rest of a series even if nothing else changed.

Where the threshold actually sits

There is no single published number, because every tool downscales to whatever its own model architecture takes as input, and none of them advertise it. What can be said with confidence is the shape of the curve rather than its exact location: a photo saved at a few hundred pixels on the long edge is almost certainly below whatever floor a modern model uses and will cost real detail; a photo from any phone's default camera setting, typically several thousand pixels on the long edge, is almost certainly well above it. The gap between those two numbers is wide enough that most people never have to find the exact threshold - staying at a phone's default output clears it by a comfortable margin, and the only way to fall below it is to deliberately shrink a file first, usually to save space before sending it somewhere.

A cheap way to check whether resolution is the culprit

If a score looks unexpectedly low and the photo in question was, for whatever reason, saved or forwarded at a reduced size, that is worth ruling out before concluding anything about the subject. Resubmit the same shot at the camera's original output size, with nothing else changed, and compare. If the number moves, resolution was doing some of the work; if it does not, the threshold was already cleared and the explanation lies elsewhere.

What this is not

This is about pixel count as it arrives at ingest, not about what happens to the file afterward - what a screenshot or a re-save does to an image is a related but separate degradation, worth its own reading. It is also not about focus: a sharp low-resolution image and a blurred high-resolution one lose detail for different reasons, and resolution alone does not explain either.

Away from the score entirely

Resolution has nothing to do with an actual measurement in centimetres, which depends on a tape and a method rather than pixel count. It matters just as much, in the same threshold way, if the image is headed to a human judge - a genuinely low-resolution photo loses legible detail for a person exactly as it does for a model, and Rate Cock's photo guide asks for a normal camera resolution for the same reason argued here.