Written by the BetterPic team. Disclosure: we sell one of the routes to getting one.
People commission a portrait and then use it as a profile picture, and are mildly disappointed without knowing why. The two are built for different viewing conditions and optimising for one costs you the other.
A portrait is looked at. A profile picture is glanced at. A portrait has a viewer who has decided to look, at a size where detail exists. A professional profile picture appears in a feed at roughly 48 pixels, beside a name, for about a third of a second. That is not a smaller version of the same job.
A portrait is seen alone. A profile picture is seen in a row. Whatever else is around it — colleagues on a team page, other candidates in a search result, other names in a thread — sets the comparison. Your photograph is judged against that context before it is judged on its own merits.
A portrait can be interesting. A profile picture has to be legible. Interesting light, an unusual crop, a meaningful location: all of these work at portrait size and turn into noise at 48 pixels.
Tonal separation between your hair and the background. This is the whole game at small sizes. If they sit at similar brightness your head loses its outline and becomes a smudge. Dark hair on mid-grey is the classic failure and it is extremely common.
Your face occupying enough of the frame. A portrait crop with generous space around the head loses the face entirely when a platform crops it square and then masks it to a circle. Crop tighter than feels right.
One dominant tone in the clothing. Patterns dissolve into visual noise. A solid mid-tone reads as a shape, which is all you get at that size.
Shrink your candidate to about 50 pixels. Put it beside three other people's from the platform it is going on. If you can still tell it is a person facing you, it works. If it has become a grey blob with a lighter patch, the problem is almost always the background rather than the photograph, and changing the background fixes it without reshooting.
Almost nobody runs this test, which is why so many otherwise good photographs fail at the size they are actually used.
You probably want two files from one session: a tight square crop for every screen use, and the original portrait framing kept at full resolution for print and for anything that wants a rectangle. Producing both takes two minutes in any editor, and it prevents the common outcome of one file being cropped badly by five different systems.
And use the same one everywhere. Recognition compounds; a face that changes every few months builds none of it. If you want to see how a generated set handles your face before paying for one, the free tier covers up to 10 images.