Some people I coach have the same awkward gap on their profile: a holiday photo cropped down to a face, or no photo at all. A session with a photographer can easily run to three figures, and when you are between jobs that is hard to justify. So clients keep asking me whether they can just turn a casual snap into a professional headshot with AI.

My honest answer used to be “probably, but be careful”. This month I tested it properly, so I can give a better one. I wanted to know three things. Does the person still look like themselves? How much does it cost and how long does it take? And where does a tidy-up turn into a misleading photo?

One point before we start. In the UK, most recruiters still prefer CVs without a photo, so this is really about LinkedIn-style profiles, speaker bios and company “meet the team” pages. If a job advert does not ask for a photo, leave it off your CV.

How I set up the test

I did not use real clients’ faces. Instead I generated two fictional people as stand-ins and gave each of them a typical “casual” photo. “Tom” is 38, sitting in a pub garden in a faded green hoodie. “Amara” is 29, smiling in her kitchen in a grey sweatshirt with a yellow headband. I gave each a few distinguishing details on purpose: Tom’s light stubble and laughter lines, Amara’s gap between her front teeth and a small mole above her lip. A real headshot has to keep exactly those things.

The first small failure came here. I asked for a pale scar through Tom’s left eyebrow, and it never appeared. Fine for a stand-in, but it shows that very small details in a prompt can be dropped.

For the editing itself I used the Nano Banana 2.1 API, an image model that takes your photo plus a written instruction and returns a new version. There is no app to download and no subscription: each image is a separate paid job, and failed jobs are refunded. You can choose 1K, 2K or 4K output and a range of shapes, so I used the 4:5 portrait format that suits most profile pages.

To measure likeness I did not rely on my own eyes alone. I ran every result through an open-source face-recognition model that scores how similar two faces are. A score of 1.0 means identical, and above about 0.36 the software treats two photos as the same person. It is a rough guide, not a verdict, but it helps keep me honest.

Test 1: three headshot styles from one casual photo

Each person got the same three requests: a plain light-grey studio backdrop with a navy blazer, a softly blurred office with a smart jumper, and an outdoor park look with a dark teal shirt. Every prompt ended with the same instruction: keep the face, hair, skin texture and expression exactly as they are, and do not slim, smooth or de-age.

All six worked first time. Each took 12 to 21 seconds and cost about 3p ($0.033). Likeness scores ranged from 0.64 to 0.82, so all were comfortably “the same person”. Looking closely, Tom kept his stubble and laughter lines. Amara kept the gap in her teeth and the mole, because I told it to.

The changes I did not ask for were small but real. In Amara’s studio version her hair came out fuller and rounder than in the original, which is the kind of thing a recruiter notices when you walk in. The headband went because I asked for her hair to fall naturally. For an interview, check the hair as hard as the face.

Test 2: cheaper route, bigger file

The model page offers two ways to run the same model. The official route is billed by usage and can return up to four images per request. The cheaper “Ext” route charges a flat price per image and returns one at a time. I ran the studio headshot both ways.

The Ext route cost $0.012 per image against roughly $0.033, took about 21 seconds, and the likeness scores were almost identical (0.74 and 0.63 against 0.71 and 0.64). I could not see a quality difference at profile-photo size. I also tried 2K on the official route: 1856 x 2304 pixels, 21 seconds, $0.047. That is more than enough for print, and far more than any profile page needs.

Test 3: editing the edit

People rarely get it right in one go. They want a tie, then a different background, then a tighter crop. Each round feeds the last result back in, and I wanted to know whether the face slowly drifts. Starting from Tom’s studio shot, I added a burgundy tie, then switched to a warm beige backdrop, then asked for a tighter crop with his eyes on the upper third of the frame.

Every instruction was carried out. The crop was accurate: his face got 26% larger and his eyes landed a third of the way down the frame. But the similarity to round one slipped a little each time: 0.97, 0.95, 0.92. That is still clearly Tom, yet the direction is worth knowing. My advice is to go back to the original photo for each new version rather than editing the edit five times.

Test 4: what over-retouching looks like

This is the test I most wanted my clients to see. I asked for a “flawless magazine-cover” version: remove every line and blemish, airbrush the skin, whiten the teeth (and close Amara’s gap), slim the jaw and take ten years off.

It did all of it. Tom lost his stubble and several years; Amara’s gap disappeared, her skin went glossy, and the headband came back from the original photo. Here is the surprise: the face-recognition software still scored them 0.71 and 0.68, about the same as the natural edits. Software that matches faces is quite forgiving. People are not. An interviewer who has looked at your photo will notice, within seconds, that you have more lines and a different smile, and the first impression you have built turns into a small question mark.

What it cost, and my coaching rules

The whole test, including making the two stand-in photos, came to 16 images, no failed jobs and $0.50 (about 40p). Most images took 12 to 22 seconds; the slowest, the first casual photo, took about a minute. Compared with a studio session, the money is not the issue. The judgement is.

These are the rules I now give clients:

  • Start from a recent, well-lit photo of you as you look now. Same hairstyle, glasses, beard and hair colour you will bring to the interview.
  • Change the setting, not the person. Background, clothes and lighting are fair game. Face shape, skin tone, teeth, weight and age are not.
  • Edit from the original each time. Small drift adds up over rounds.
  • Look at it next to a mirror. If a friend would say “that’s a good photo of you”, keep it. If they say “you look different”, bin it.
  • Be ready to say how it was made. If someone asks, a relaxed “I tidied up a casual photo with an AI tool” is fine. Pretending it was a studio shoot is not.
  • Never use someone else’s face, and do not use AI images for ID checks or right-to-work documents.

Used that way, an AI headshot is a perfectly sensible way to fill an empty profile picture for the price of a stamp. Just remember who the photo is for. The person reading your profile will meet you in the end, and the photo should help them recognise you, not surprise them.

 

Comments

comments

Pin It on Pinterest

Share This