This Photo Will Be Shown to Most People at 48 Pixels

How it works

The upload is scaled to about one and a fifth megapixels and repainted at the lowest denoise in this family, because the instruction changes less than the others do. It asks for three things: crop in so the head fills most of the frame, separate the person from the background by tone, and keep the face sharp and evenly lit. Features, expression, and clothing are all held.

After decoding, the result is scaled to a fixed 1024 by 1024 with a center crop, so what you download is already square.

The negative prompt is written in the vocabulary of the failure rather than of the picture: head too small, cluttered background, low contrast against the background, cropped chin.

Design for the size it is actually seen at

The instinct with a profile picture is to judge it at the size you are looking at it, which on your own screen is large.

That is the wrong size. Count where impressions happen — a comment, a feed post, a suggested connection, a notification, a search result — and nearly all of them render a small circle. Your profile page, where the picture is big, is seen by a fraction of the people who see the thumbnail.

So the rule is to make the decision at forty-eight pixels. Scale the candidate down in any viewer and look at it. Almost every question resolves immediately at that size, and questions that only exist at full size were never load-bearing.

Head size and contrast are the whole game

Downscaling is averaging. Take a hundred pixels of eye, cheek, and hair and you get one pixel of their mean, so anything smaller than the sampling grid stops existing.

What survives is large and high-contrast. The outline of a head against a different tone survives. A dark jacket against a dark room does not — it merges into one mass, and the thumbnail reads as a shape rather than a person. A head occupying a fifth of the frame becomes ten pixels of face, which is not enough for anyone to tell you from anyone.

That is why this page's two instructions are crop and tone separation rather than anything about polish. They are the two properties that survive the resize. The tone separation is achieved by changing the background, which is worth knowing: if your background is meaningful to you, this is not the page for it.

What it will not do

It will not add what the frame does not contain. If your chin sits at the very bottom edge of the source, cropping in has nowhere to go, and the pass either keeps the tight frame or invents jaw — both bad outcomes, and the second is worse because it looks fine.

It will not fix focus. A soft face is soft, and at thumbnail size that matters less than you would think, which is the one consolation this page offers about bad photos.

And it does not make a headshot suitable for print, a directory, or anywhere the image is shown large. It optimizes for a small circle, and that optimization — crop in hard, simplify the background — is genuinely wrong for a page that shows the picture at full width.

Publication gate

This page ships once the workflow has been run against a half-body conference photo, a dark-haired subject against a dark wall, a portrait with the chin near the bottom edge, and a photo already cropped tight, with every output inspected both at full size and downscaled to forty-eight pixels.