How to Create the Perfect AI Headshot for LinkedIn

A LinkedIn headshot has one job: make you recognisable to someone who is deciding whether to reply. Everything else — the lighting, the background, the crop — exists to serve that. "Perfect" is not a flattering photograph. It is a current, legible likeness that survives being shrunk to the size of a fingernail.

That framing matters because it tells you what to ask an AI tool for. Used narrowly, to fix lighting, tidy a background, or rescue a photo taken in bad conditions, these tools produce a better headshot than most people can take on a phone. Used broadly, to "make me look professional", they drift your face and produce something that reads as synthetic to exactly the audience you were trying to impress.

Start from a photo that already looks like you

Nothing downstream recovers a bad source. The model works from what you give it, and every ambiguity in the input becomes an invention in the output.

A usable source has four properties:

• Your face is unobstructed. No sunglasses, no hand near the chin, no hair across one eye. Anything the model cannot see, it will guess at.
• The light is even. Soft light from a window, with your face turned slightly toward it, beats a ceiling light directly overhead. Harsh overhead light puts shadows in the eye sockets, and models fill those shadows with invented structure.
• It is recent. Within two or three years, or sooner if your hair, glasses or facial hair have changed.
• It is sharp at full size. Open it at 100 percent. If the eyelashes are mush, the model has nothing to sharpen and will hallucinate detail instead.

The single most useful thing you can do is retake the source. Stand a metre from a window on an overcast day, have someone take three or four frames on a phone at eye level, and pick the sharpest. That takes five minutes and improves the result more than any setting.

If your only option is an older or lower-resolution photo, upscaling it first gives the headshot step more to work with. Do the two operations separately rather than asking one prompt to do both.

Give the edit a narrow job

The difference between a headshot that works and one that looks uncanny is usually the size of the instruction.

Narrow instructions that tend to work:

• "Replace the background with a plain neutral grey studio backdrop."
• "Even out the lighting on the face, keep the existing skin texture."
• "Remove the lanyard."

Broad instructions that tend to fail:

• "Make me look like a successful executive."
• "Professional headshot, magazine quality, perfect skin."

The second kind gives the model licence to change your face, because "successful" and "perfect" are not visual instructions — they are judgements, and the model resolves them by moving toward an average face it has seen labelled that way. That average is younger, more symmetrical, and less like you.

In practice: change one thing per run. If you want a new background and better lighting, do the background, check it, then do the lighting. You will spend one extra generation and get a result you can actually verify.

Frame for the crop LinkedIn will apply

LinkedIn displays your photo as a circle. It is shown large on your profile, small in search results, and around 40 pixels across in the feed and in comment threads. Most headshots are composed for the large version and fall apart in the small one.

Two rules cover it:

Your face should fill roughly 60 percent of the frame height. Head and the top of the shoulders. Not a full torso, not a tight crop that clips your forehead.

Your eyes belong about a third of the way down. This is the oldest rule in portrait composition and it is the one that survives aggressive downscaling, because the eyes are what the viewer's own recognition system looks for first.

Then check the circle. Square up the image, imagine the inscribed circle, and confirm nothing important sits in the corners that will be cut. A shoulder logo or a chin that touches the bottom edge both disappear in the crop.

Run three checks before you publish

These take about a minute and catch almost everything.

The likeness check. Open the source and the result side by side at full size. Look specifically at the distance between the eyes, the shape of the nose bridge, the hairline, and the jaw. These are where drift shows up first and where you are least likely to notice it on a single image. If a colleague would hesitate for a second, regenerate.

The thumbnail check. Shrink the result to about 40 pixels wide and look at it. Does it still read as a person, and as you? A photo that only works at full size is a photo that does not work on LinkedIn.

The artifact check. Zoom to 100 percent and check the places models get wrong: the boundary between hair and background, the frame of your glasses where it crosses your temple, teeth, ears, and any text on clothing. Hair against a replaced background is the most common failure — look for a halo, or for individual strands that end in nothing.

What a good result actually looks like

It looks like a photo of you on a good day, taken by someone who knew what they were doing. Skin has texture. There is a shadow under the jaw. The background is plain but not flat. Nothing about it announces itself.

If your reaction is "that is a nice photo of me", it worked. If it is "that looks amazing", look again — you are probably looking at a face that has drifted toward the average, and the people who know you will see it even if you do not.

Where the honest limits are

AI headshots are good at lighting, backgrounds, and small distractions. They are unreliable at anything involving fine repeating structure — glasses frames, jewellery, patterned fabric, text — and they have a documented bias toward smoothing skin and lightening skin tones. None of that makes them unusable. It means you check the output rather than trusting it.

They also cannot fix a photo that does not exist. If you have never had a photo taken where you look approachable, no model will invent one that is still you. Take the five minutes by the window first.

A workable sequence

1. Take three or four fresh frames near a window at eye level, or pick the sharpest recent photo you have.
2. If it is soft or small, upscale it as a separate step.
3. Run one narrow instruction — usually the background.
4. Check likeness, thumbnail, artifacts.
5. Run a second narrow instruction if you need it, and check again.
6. Crop to a square with your eyes a third down, and upload.

Four of Kitana's tools run in the browser, so you can work through this without installing anything — and a background swap and a headshot pass are two of them. If you want the rest of the toolkit, including restoration and object removal, the iOS and Android apps carry all thirteen.

The related question of whether recruiters mind is worth its own answer, and we have written it up in AI headshots for job applications. The short version: they mind drift, not AI.